Subscribe
Share
Share
Embed
On this week's show, Adam Leventhal posed questions from Hacker News (mostly) to Oxide founders Bryan Cantrill and Steve Tuck. Stick around until the end to hear about the hardest parts of building Oxide--great, surprising answers from both Bryan and Steve.
Questions for Steve and Bryan:
[@6:38] Q:
Congrats to the team, but after hearing about Oxide for literal years since the beginning of the company and repeatedly reading different iterations of their landing page, I still don't know what their product actually is. It's a hypervisor host? Maybe? So I can host VMs on it? And a network switch? So I can....switch stuff? (*)
A:
Steve: A rack-scale computer; "A product that allows the rest of the market that runs on-premises IT access to cloud computing."Bryan: agrees
[@8:46] Q:
It's like an on prem AWS for devs. I don't understand the use case but the hardware is cool. (*)I didn’t understand the business opportunity of Oxide at all. Didn’t make sense to me.However if they’re aiming at the companies parachuting out of the cloud back to data centers and on prem then it makes a lot of sense.It’s possible that the price comparison is not with comparable computing devices, but simply with the 9 cents per gigabyte egress fee from major clouds. (*)
A:
Bryan: "Elastic infrastructure is great and shouldn't be cloistered to the public cloud"; Good reasons to run on-prem: compliance, security, risk management, latency, economics; "Once you get to a certain size, it really makes sense to own"Steve: As more things move onto the internet, need for on-prem is going to grow; you should have the freedom to own
[@13:31] Q:
Somebody help me understand the business value. All the tech is cool but I don't get the business model, it seems deeply impractical.So you have to be this ultra-bizarre customer, somebody who wants their own servers, but doesn't mind VMs, doesn't need to migrate out of the cloud but wants this instead of whatever hardware they manage themselves now, who will buy a rack at a time, who doesn't need any custom hardware, and is willing to put up with whatever off-the-beaten path difficulties are going to occur because of the custom stuff they've done that's AFAICT is very low value for the customer. Who is this? Even the poster child for needing on prem, the CIA is on AWS now.I don't get it, it just seems like a bunch of geeks playing with VC money?(*)
- You buy your own servers instead of renting, which is what most people are doing now. They argue there's a case for this, but it seems like a shrinking market. Everything has gone cloud.
- Even if there are lots of people who want to leave the cloud, all their data is there. That's how they get you -- it costs nothing to bring data in and a lot to transfer it out. So high cost to switch.
- AWS and others provide tons of other services in their clouds, which if you depend on you'll have to build out on top of Oxide. So even higher cost to switch.
- Even though you bought your own servers, you still have to run everything inside VMs, which introduce the sort of issues you would hope to avoid by buying your own servers! Why is this? Because they're building everything on Illumos (Solaris) which is for all practical purposes is dead outside Oxide and delivering questionable value here.
- Based on blogs/twitter/mastodon they have put a lot of effort into perfecting these weird EE side quests, but they're not making real new hardware (no new CPU, no new fabric, etc). I am skeptical any customers will notice or care and would have not noticed had they used off the shelf hardware/power setups.
A:
Bryan: "EE side quests" rant; you can't build robust, elastic infrastructure on commodity hardware at scale; "The minimum viable product is really, really big"; Example: monitoring fan power draw, tweaking reference desgins doesn't cut it Example: eliminating redundant AC power suppliesSteve: "Feels like I’m dealing with my divorced parents" post
[@32:24] Q (Chat):
It would be nice to see what this thing is like before having to write a big checkSteve: We are striving to have lab infrastructure available for test drives
[@32:56] Q (Chat):
I want to know about shipping insurance, logistics, who does the install, ...Bryan: "Next week we'll be joined by the operations team" we want to have an indepth conversation about those topics
[@34:40] Q:
Seems like Oxide is aiming to be the Apple of the enterprise hardware (which isn't too surprising given the background of the people involved - Sun used to be something like that as were other fully-integrated providers, though granted that Sun didn't write Unix from scratch). Almost like coming to a full circle from the days where the hardware and the software was all done in an integrated fashion before Linux turned-up and started to run on your toaster. (*)
A:
Bryan: We find things to emulate in both Apple and Sun, e.g., integrated hard- and software; AS/400Steve: "It's not hardware and software together for integration sake", it's required to deliver what the customer wants; "You can't control that experience when you only do half the equation"
[@42:38] Q:
I truly and honestly hope you succeed. I know for certain that the market for on-prem will remain large for certain sectors for the forseeable future. However. The kind of customer who spends this type of money can be conservative. They already have to go with on an unknown vendor, and rely on unknown hardware. Then they end up with a hypervisor virtually no one else in the same market segment uses.Would you say that KVM or ESXi would be an easier or harder sell here?Innovation budget can be a useful concept. And I'm afraid it's being stretched a lot. (*)
A:
Bryan: We can deliver more value with our own hypervisor; we've had a lot of experience in that domain from Joyent. There are a lot of reasons that VMware et al. are not popular with their own customers; Intel vs. AMDSteve: "We think it's super important that we're very transparent with what we're building"
[@56:05] Q:
what is the interface I get when I turn this $$$ computer on? What is the zero to first value when I buy this hardware? (*)
A:
Steve: "You roll the rack in, you have to give it power, and you have give it networking [...] and you are then off on starting the software experience"; Large pool of infrastructure reosources for customers/devs/SREs/... in a day or less; Similar experience to public cloud providers
[@01:02:06] Q:
One of my concerns when buying a complete solution like an iPhone (or an Oxide rack 😉) is how much ownership I actually have over the device. Where does Oxide draw the line on "ownership" of the rack? (*)
A:
Bryan: "You bought the computer, it belongs to you"; it is important to us to enable that ownershipSteve: operator level and developer level; "Anything that we are injected ourselves into is [...] in the interest of making sure that the customer is getting a highly available and efficient system"
[@01:05:03] Q (Chat):
Do you sell support contracts?
A:
Steve: Yes, comes with a support subscription (typically three years, flexible)
[@01:05:37] Q (Chat):
Is the rack dependent on any Oxide cloud services?
A:
Bryan: "Emphatically not" The customer has total agency over data I/OAdam: Updates will be fetches from a web server
Interlude
[@01:07:49]
This is super cool. I realize a lot of HN folks might not see the point of this, but it literally saves an entire team of people for companies. (*)
[@01:08:00]
Oxide is such an ambitious project, I am such a fan of the aesthetic and design and of course transparency of all of the cool people that work there. (*)
[@01:09:44]
FYI for language pedants like me: It’s “on-premises” or “on-prem”. A “premise” is something assumed to be true. (*)
[@01:11:43] Q (Chat):
When is oxide gonna make a smaller rack?
A:
Bryan: Probably no product for the homelab; "Architecturally, there is nothing here that won't scale down'; ear to the marketSteve: can be scaled down from a power perspective
[@01:21:53] Q:
Are you shipping Milan racks and if so what does the moving to Genoa entail for Oxide? (*)The listed items are for last gen CPUs and 2 gen old nics (100 Gb/s). Also zfs rather than ceph. It's hard to consider under those constraints. Any plans for upgrades beyond single rack deployments and newer gen hardware? And what is the sell case over kube, mesos, or VMware, especially since the hardware and software appear tightly coupled? (*)
A:
Bryan: Keeping a close ear to the market; "You will see a Genoa-based product"Steve: You can add next-gen sleds to the rack, you have modularity
[@01:26:10] Q:
Do you have any liquid cooling solutions collaborators? (*)
A:
Steve: Collaboration with Motivair; The product is conventionally cooled right now. Looking at what makes sense in future architecturesBryan: Rack power budget is a challenge
[@01:28:41] Q:
Why do you use Illumos? (*)
A:
Bryan: "We need to be able to control our fate"; dtrace;
Adam: "It was our fastest path to market (both development and support) with our team, expertise, and experience."
[@01:41:20] Q:
Congrats! This seems to be a cool product! How does this work with software updates? Are they subscription-based, or are they for free forever? In case the company goes bankrupt, will the software still work, and is the source code open? (*)
A:
Bryan: "In many ways we viewed the ability to upgrade as the most important feature" The software is open by default, and we're reviewing instances where we're constrained by IP we don't own.
What part of this journey was hardest? E.g., the scope increase when you couldn’t use reference designs? 😉 Dealing with the supply chain? Raising money for this kind of startup? (*)
A:
Bryan: "when we got funding" When you have a blank sheet of paper, where do you start?Steve: SVB collapse and associated banking crisis
In addition, there was a live chat during the show where members of the Oxide team answered additional questions and in greater depth.
If we got something wrong or missed something, please file a PR! Our next show will likely be on Monday at 5p Pacific Time on our Discord server; stay tuned to our Mastodon feeds for details, or subscribe to this calendar. We'd love to have you join us, as we always love to hear from new speakers!
Creators & Guests
What is Oxide and Friends?
Oxide hosts a weekly Discord show where we discuss a wide range of topics: computer history, startups, Oxide hardware bringup, and other topics du jour. These are the recordings in podcast form.
Join us live (usually Mondays at 5pm PT) https://discord.gg/gcQxNHAKCB
Subscribe to our calendar: https://sesh.fyi/api/calendar/v2/iMdFbuFRupMwuTiwvXswNU.ics
Hello, everyone. First of all, Steve is with me here. Steve, hello. Hello. I we we'll both appear as my voice.
Speaker 1:We so, very exciting because we launched on on Friday. We got we are, the the first oxide rack came off the line and was was crated up, and, it was really exciting, today in the office. In fact, you may hear us chewing a little bit here, Adam. There you go. As as Steve shakes a, a a package showed up in the office today from nuts dot com.
Speaker 1:Mhmm. Like, and, like, did you oh my god. What did we accidentally order something, like, nuts dot com? It's a
Speaker 2:good size box. Pretty heavy.
Speaker 1:Good size box. How heavy would you say it was? It was on the order of about £5. It was actually £7 according to this invoice that I have.
Speaker 3:How many gift cards is that? That's their question.
Speaker 1:That that's exactly right.
Speaker 2:It's like Christmas morning. Right.
Speaker 1:And inside, and this is, you know, what we have we have said from the beginning that, one of the, terrific things that we have enjoyed as a company is the number of folks out there who've been really supportive of what we've been doing. And, it's really meant a lot to us, even those folks that we know that for whatever reason aren't gonna be aren't necessarily Oxide customers, but they've been supporting us from afar, and that's been really meaningful for us. And indeed, this package from nuts.com was from one such supporter with the gift message, way to go. Congrats to you all. This is for everyone who thought you were nuts.
Speaker 1:So thank you very much. You know who you are. We really appreciate
Speaker 2:it. Is that end of message?
Speaker 1:It did say, this is not end of message. If you'd like me to, I will read the end of the message. Just curious. Okay. It's a little more barbed.
Speaker 1:This is for everyone who thought you were nuts, Emdash. Fuck them. I guess that's what you were you really wanted me to get to them. I know. I know.
Speaker 1:I know. The, you're right. No. Thank you very much. And, Yeah.
Speaker 1:We did tell we did have definitely some people who thought we were, a little bonkers, but, we are, very, very excited to hit this milestone. Of course, we've got, a long way to go from our perspective in terms of all the things that we wanna go build out on on top of this, and we're super excited. And, you know, I I was even I get I was kind of debating whether we're gonna tweet something out on Friday. I guess we we I mean, you and I were kind of, I don't know. Maybe it was never a debate in your mind whether we should tweet something out as a company.
Speaker 1:But Yeah. I mean,
Speaker 2:it felt like yeah. And after after this long of a road, And it it it was a bit surreal on Friday to go from where we knew we were gonna be, but to actually have that system loading on
Speaker 1:a truck and and shipping out. It was surreal. And in part because we've been working so hard for so long, and I really did liken it to, to a summit. When you're when you're climbing a mountain and you have so many false summits, and you just get to the point where you're just, like, grinding, and then all of a sudden you'd be surprised, but, like, oh god. Like, that's that's the summit.
Speaker 1:We've actually got this thing, got this thing done. So it was been, of course, again, we got a lot of a lot of work to do and a lot of, lot of software that we will continue to evolve, but, but the but the hardware is, is done. They're they're they're all exciting.
Speaker 2:A lot of people out there that were waiting and waiting and waiting and waiting. And so I think just as a way to to note that we had begun shipping commercial systems, It was worth a quick mention.
Speaker 1:And and there were some naysayers out there. There were some naysayers. I I you know, I I don't think I'm going to this person deleted their tweets, so I'm not gonna name them. But there is someone who sue said they were gonna eat their shoe if if Oxide ever shipped anything.
Speaker 4:I think we've been holding on to that. I mean, 3 years ago, 2 years ago, I mean, we've been holding on to it. Like, a a tweet They may they may have deleted it. All of this to us. Yeah.
Speaker 4:Yeah.
Speaker 1:They may have deleted it. It's true. Yeah. I know. Adam, we have we kind of, like, put it up in the locker room, you know, as the the this is,
Speaker 2:To their credit, they came back forward and had noted it and and is are now offering up a case of beer.
Speaker 1:Yeah. And, like, not to us. They were talking about, like, hey. I know at at this conference, I was saying these guys weren't gonna ship Very rarely.
Speaker 2:They were saying it all over.
Speaker 1:They were just saying it all over the place. A lot of checks were written. Exactly. But, that was fun. So, I think the Adam, we're talking about spending today going through the so the Hacker News thread, was filled with a lot of delightful support.
Speaker 1:Also, some questions. And then also just because because I, called out Hacker News for what a supportive community is, of course, Hacker News has gotta bring has gotta be the most orange site, and, there were definitely some comments that were little bit trollish. We'll try not to spend too much time on those, but we also can't get out of here without getting some of those.
Speaker 4:So That's true. Yeah. I thought I thought I'd go through some of the questions in an order that seemed reasonable to me. I do wanna pause and then just say, you know, I know congratulations to us all, congratulations to the Oxide team, but I feel like I'd be remiss to not saying congratulations in particular, Brian and Steve. Because, like, you started the thing, you had the vision, you raised the money, you assembled the team.
Speaker 4:So, I still wanted to go unsaid that you guys should feel a tremendous amount of pride in what we, the whole team, have put together. But, but thanks to you guys.
Speaker 1:I think that, you know, yeah, I appreciate it, but I think you can see speak on behalf of Steve that, like, it feels it is such a comprehensive team effort that it's almost like
Speaker 2:it I mean just saying today, it feels like we have a colleague of ours who started in January, and it's felt like she's been here
Speaker 4:for
Speaker 2:the whole the whole journey.
Speaker 1:The whole journey. Yeah. I I feel yeah. It it it
Speaker 2:It's hard to remember for most folks on the team, like, that they weren't there from the very beginning, which Yeah.
Speaker 1:Right. In part because we needed every single one of them to ship this thing. They're like because we would have been completely screwed without every single one of them. So, Yeah. So Yeah.
Speaker 1:No. Thank you. But Adam you, though. Thank you. And, it's it's been it really been exhilarating.
Speaker 1:It's a lot of fun.
Speaker 4:So, there's a general category of kind of what is oxide.
Speaker 1:And I
Speaker 4:hope you guys have been to a seder, but one of the traditions in a seder is or one of the parts of a seder is talking to the story of the 4 children. One of the children is the one who doesn't know how to ask the question, so I'm gonna ignore that child. But there are there are 3 other children. There's the simple child who asks the question in a simple way. There's the wise child who asks the the question about, you know, what is this in a more sophisticated way.
Speaker 4:And then there is the wicked child. So we're gonna start with the simple child, because we got a comment, saying congrats to the team after hearing about Oxide for years since the beginning of the company repeatedly reading different iterations of the landing page. I still don't know what the product actually is.
Speaker 1:It's
Speaker 4:a hypervisor host, maybe, so I can host VMs on it and a network switch so I can switch stuff. So why don't we start there?
Speaker 1:What is this thing anyway? Yeah. Yeah. What do we build?
Speaker 2:So it it it's a a question we've definitely gotten a lot because for a while, we had a couple pictures on a website that were all hardware centric. And folks had a sense we're building some sort of a computer, but that was about it.
Speaker 1:Or we're a firmware company. That's the other thing you get a lot of. It's like, are you just doing firmware? It's like, well, yes. And?
Speaker 2:Yeah. Or I I think in particular VC that was, like, in a you know, if you're doing firmware Right. You might wanna invest. And the difficulty is that we're doing we're doing probably 8 different kind of products, designed together into 1. But but, fundamentally, what we're aiming to go build is a a a product that allows the rest of the market that runs on premises IT access to cloud computing.
Speaker 2:And that's what we're that's what we're we were aiming for from the beginning is we are big believers in cloud computing. We think about cloud computing not as renting someone else's computers, but as having abstractions that allow you to consume large pools of infrastructure resources to deploy and run software. And the the kinds of systems that cloud hyperscalers have in their own data centers look nothing like the hardware and software that folks that are running on prem IT have access to. And so we are building a computer. This computer is at rack scale, and it comes complete with the innovations that the hardware hyperscalers have built for themselves and a complete software set that allows you to expose cloud computing services to your users on prem.
Speaker 1:You know, that Our project is great. Together.
Speaker 4:That's a great segue to what I'll what I'll describe as the wise child. I got a couple of these. One says, it's like on prem AWS for devs. I don't understand the use case, but the hardware is cool, says one. I noticed this, I don't understand the business opportunity for Oxite at all.
Speaker 4:Doesn't make sense to me. However, if they're aiming at companies parachuting out of the cloud, backed data centers, and on prem, then it makes a lot of sense. It's possible that the price comparison is not comparable with computing devices simply anyway, and so on. So, I think a a lot of the premise there being isn't everyone in the cloud?
Speaker 2:Yeah. Why would anyone run anything outside of the public cloud?
Speaker 1:Isn't Jeff Bezos gonna own and operate every computer on the plan? And our belief, of course, is no. Is that elastic infrastructure is great. And just as Steve said, that shouldn't be cloistered to the public clouds. And there are lots of good reasons to run on prem.
Speaker 1:And in fact, we're so deep into the public cloud revolution that most of the bad reasons to run on prem have evaporated. And if you are, I you know, there is there's something kinda windy about the to to use an overly trendy term about those that are on prem, that if you're running on prem in 2023, there are probably some good reasons for it. And those good reasons are around, regulatory compliance, or they're around security, or they're around risk management, or they're around latency, or, and this is kind of what the the comment was getting to, they're around economics. And as it turns out, if you're gonna run a lot of compute, you may wanna own some of it, as opposed to renting it all the time. And, you know, being in the in the public cloud is is, you know, kind of in an Airbnb.
Speaker 1:And, you know, there there can some upsides in Airbnb, but, or you're renting an apartment and, you know, maybe the Airbnb maybe Airbnb is like, is that Lambda, or is that the hotel room? So we got the the hell. Yeah. Right. Room service with with Lambda, and then you're kind of but you ultimately, you know, you're renting can make a lot of sense.
Speaker 1:But once you get to a certain size, it really makes sense to own. And that's that's our big belief.
Speaker 4:Yeah. And to Steve's point earlier, the the cost of owning is, you know, ex explodes with all the operational cost that the developers then have to onboard where they don't get to just provision VMs with an API. They have to, like, file tickets and so forth.
Speaker 1:That's right. Exactly it.
Speaker 2:Yeah. And I think, I I mean, just to to drive the point home, you've got, I think it was Andy Jassy, who at the time was running AWS, was on stage at re Invent in 2021. And he was describing the opportunity for AWS, which in 2021, I think everyone, certainly just ask any VC, everyone was running in the public cloud or was gonna be running in the public cloud shortly. And Andy had said that, 95% of IT infrastructure was running outside of public cloud. To underscore the opportunity for AWS, I think what it highlights is just how big the on premises IT infrastructure footprint is and that public cloud is gonna grow very, very large continue to grow very, very large.
Speaker 2:The the need for on premises IT structure is going to remain very big, and as more things move onto the Internet, continue to get larger itself.
Speaker 1:And, you know, it, and TriNet has got a good, comment in the chat here, Adam, saying, hey. The way my colleague is asking, so does this mean that we can actually have access to this kind of compute in Eastern Europe or Asia? And we absolutely I mean, I think that's the, you know, whether you wanna put that as kind of regulatory compliance because certainly there are a lot of of in country regulations around data movement and so on, or often just latency. It's like, actually, you know what? I'd rather not hit US east from from, well, you know, with, from across an ocean or what have you or so there are a lot of reasons why one wants to have your own local compute in country, and we absolutely see that.
Speaker 2:And, again, we are our belief is purely that you shouldn't have to make these trade off decisions.
Speaker 4:Yeah.
Speaker 2:If if cloud if the benefits of cloud computing, namely faster, easier access for developers to get their jobs done and lower overhead for operators to run highly available infrastructure, Like, if if those are benefits that are shared, which we have come across kind of everyone, big companies, small company, do share those the the desire for those benefits, then you should have the freedom to choose. Do I wanna rent it, or do I wanna own it? Do I wanna run it in someone else's data center or run it my own data center? And it's kind of a dichotomy right now that that that you have to make this trade off.
Speaker 4:So that takes us to what I'm gonna characterize as the wicked child here. So another comment, and and it's pretty long, so I'm not gonna read it, in its entirety, but there's lots to unpack as this. He says, somebody help me understand the business value. All the tech is cool, but I don't get the business model. It seems deeply impractical.
Speaker 4:Based on blogs and Twitter and Mastodon, they put a lot of effort into perfecting these weird EE side quests Mhmm. Without making real new hardware, no new CPU, no new fabric, etcetera. I am skeptical any customers will notice or care and would have not noticed had they used off the shelf hardware power setups. So you have this ultra bizarre customer.
Speaker 1:Is it this is the Wiggin Road. This is the
Speaker 4:Someone someone who wants their own servers but doesn't mind VMs, doesn't need to migrate out of the cloud but wants this instead of whatever hardware they manage themselves now, Who will buy a rack at a time, who doesn't need any hard custom hardware, and willing to put up with whatever off the beaten path difficulties are going to occur? Who is this? Even the poster child of meeting on prem, the CIA is on AWS now. I don't get it. Just seems like a bunch of geeks playing with VC money.
Speaker 1:Oh, boy. I even you said, like, I'm not gonna read this whole thing and then How long was it?
Speaker 4:Yeah. I feel like you Oh. Oh, there's more It
Speaker 3:was longer. It was longer.
Speaker 1:It was longer than that. Okay. So I I mean, boy, where do you wanna start?
Speaker 4:I the first one,
Speaker 1:just because yeah. Sorry.
Speaker 3:The the funniest thing, I think, to me about this, I I appreciate skepticism, but when people come at it from a, like, therefore, I know that no one will buy this is very different than, like, I don't understand this because I'm not in this space. Like, I think a lot of people are skeptical about the business model because they've just never dealt with this area before, and so, you know, they naturally are, like, confused. But I think the funniest thing about this comment is it's on a picture of someone who bought a rack from us. So being like, no one no customers could possibly exist in response to, like, a photo of a customer having purchased a thing is just kind of, like, inherently a little funny to me.
Speaker 1:Yeah. Right. But so there's that. There is the and, you know, I I did kinda blow a gasket on EE side quests, Adam, on this
Speaker 4:I thought about digging that out. I'd be liking it.
Speaker 1:No. You can't. We you gotta I I and, you know, I think the double e's maybe had a better sense of humor about this than I did. I I definitely, we all as a team have worked really, really hard. But the double Es have been, I mean, have been extraordinarily so.
Speaker 1:I mean, again, everyone has worked really hard on this, but, boy, if there's any team to accuse of going on side quest, it is like, there've been no like, what are we talk and I did have, like, I did have a follow-up comment. I did, like it was funny because, like, in the internal chat, I'm like, I really have to reply to this comment. And immediately, people feel it's, like, their responsibility to be, like, why don't you say the comment in here where it's safe as opposed to like, I don't know what people think I'm I'm gonna say. Right. Right.
Speaker 1:Right. Exactly. I feel like this is to possible reply to that, which is really just like, what a double e side quest are we talking about? There have been no double e side quests. I do think that, it is really, really hard to go build all of this.
Speaker 1:And we have been talking about all of the and we have believed strongly in being transparent about all of the details in building this thing. I think I think they're extraordinarily technically interesting. I think that they are broadly hidden. People don't talk about them. That's a part of the reason we talk.
Speaker 1:And so it's like, no. Sorry. The the EEs are not on a side quest. You yourself, to your listener, may be on a side quest when you listen to us talk about them. It may be for you a side quest, and it may be for you interesting to hear about something, just has been interesting for me and Steve for you and Adam for you, Dave, to learn about these aspects of the system that we just really hadn't thought about before, but all of it has been actually essential for delivering the artifact.
Speaker 1:I just really wanna be unequivocal about that.
Speaker 2:Well and and it's, for a bunch of these, it's like, oh, well, just go pose that same question to hyperscaler. It's like, oxide, why'd you build your own switch? It's right. Well That double e side quest.
Speaker 1:That double e side quest. Right.
Speaker 2:It's like And then you go, you know, and and and folks from AWS will talk about, you know, building their own switch, being seminal of them actually kind of turning the corner for running through cloud infrastructure services, and that's bad thing.
Speaker 1:Yeah. And I think that, you know, I and this also gets, Adam, to, you know, part of the origin for a bunch of us here, especially coming from the software side in terms of, like, why and and maybe one way to kind of, phrase the wicked child's question in the most generous possible way is, why can't you just do this on commodity hardware? And, trust us, we tried. I mean, that's what we did. That's our life before oxide is trying really, really hard actually to do it on commodity hardware.
Speaker 1:And and, you know, Adam, you you and I were at Sun back in the day, and we you know, where we really believed in this kind of full hardware software integration. And then we kinda you know, if you talk about a side quest, I I spent, you know, 10 years, 9 years at a public cloud computing company attempting to build reliable robust elastic infrastructure on commodity hardware. And you can kinda do it a little bit, but you can't do it at scale, and then there are a whole bunch of things that you just cannot do. And you actually and the the the belief, kind of the the fundamental technical belief that we came to was we actually need to have both hardware and software together to fully deliver true elastic infrastructure to the end user. And I gotta say, now, three and a half years into this, having built this thing, I feel absolutely unequivocal on that.
Speaker 1:I mean, that was I I would say we had a pretty strong hunch going in, pretty strong belief going in, but, believe that more strongly than ever that you need both hardware and software together to deliver infrastructure.
Speaker 4:Yeah. Yeah. Brian, I I've been at arm's length from a lot of this stuff and and exposed a lot of it, on this show. But is it fair to say a lot of it was kind of all or nothing? That is to say, we couldn't sort sort of have a root of trust and an a speed, BMC or something like that that,
Speaker 1:you know, there wasn't sort
Speaker 4:of a way to split the baby on this one.
Speaker 1:There there was no way to split the baby. And then this is, like, actually a big challenge we have had just as a company is what is the minimum viable product? I mean, we're a start up. You wanna actually have you wanna have the minimum viable product. Like, that's a good idea.
Speaker 1:The problem with this is that the minimum viable product is really, really big. And, Adam, you're exactly right. There's no way to and, I mean, there have been companies that have tried to deliver just a slice, and it's really hard to succeed for reasons that are because you only delivered a slice of what you need, and no one wants to buy the slice. They wanna buy the full solution. So it is, you really need to do the whole thing.
Speaker 1:And I think, you know, one of the things one of the big lessons that we had, early on is you would kinda had this idea that we'd be able to and I, again, I hate bringing this up again because I just so hate the number of times I said this aloud that we are going to quote, unquote tweak reference designs. And the You did say that kind
Speaker 4:of a lot.
Speaker 1:Okay. You really did not need to back me up on that. I mean, really could've sat that one out. I mean, you could've been I just did not need the yeah. No.
Speaker 1:I know. I said it a lot. I said it a lot, and I I maybe I don't know. Maybe I was I guess I was hopeful. But the once you I mean, just the act of getting rid of the BMC, reference design is out.
Speaker 1:Like, you every single reference design that that ASP BMC is so entrenched into every reference design, that the second you get rid of that thing, like, yeah, you're on your own. And you're on your own, and you're in by the way, nobody does it that way. You're in absolute like bonkers territory. And, but boy, again, having done it now, there is so much, I mean, we control the system to such a deep level, and there are things that are just super natural for us, super space natural, that that are very natural for us.
Speaker 4:There's so much supernatural about the way it's supposed to work.
Speaker 1:And poltergeists in the system that we have yeah. No. The the there are so many things that are very natural for us. So I mean, as an example, I had a, I was talking to a someone who deploys a lot of infrastructure, a lot of physical infrastructure. And, in particular, he was like, can you just tell me that you can monitor the power draw on the fans?
Speaker 1:Like, sure. Absolutely. We know exactly how much I mean, for we know that actually in a couple of different dimensions. But, yes, we absolutely know how much the fans are drawing. It's like, okay.
Speaker 1:Thank god. Because let me describe this bug I had for you where a a cascade of firmware bugs from HPE in this case, but it could be easily Supermicro or Dell. But I had a a cascade of bugs where the the the iLO, which is their BMC, was mistakenly was using the draw of the CPU to determine what to do with fan speed, and and for other bugs, basically, it had to do that. And he had a spiky workload, and what would happen is the workload would spike, and they would see the current in rush. And they would do what feels like the conservative thing, which is crank the fans, and then they would kinda slowly decay that fan speed.
Speaker 1:And the problem is that the workload would spike with some regularity. Not enough regularity to actually really seriously warm up the part, but enough regularity for for the Ilo to crank the fans and leave them cranked. And what he talk about a side quest. What he had gone on was, like, I just wanna determine how much power is going into the fans. That's it.
Speaker 1:And he ultimately got to the answer, which was galling, and he was across his entire infrastructure. It was, like, 100 of kilowatts that he was burning on this particular bug. The, but it was excruciating to just determine the draw of the fans. And this is the kind of example where that decision is made so far away from the system software that actually has insight into what's going on from a workload perspective, that there's no way to just, like, tweak a design to answer that question. You really need to kinda start with a clean sheet of paper, and that's the kind of thing that we've been able to do that is easy for us.
Speaker 1:I mean, it is easy for us to tell you exactly where the draw is going, because we, at every juncture, we have picked components that allow us to write the software, that allow us to query it, to understand what's going on, and ultimately correlate that with what you're actually running in the hypervisor, in the control plane. We can actually get you to that highest level question, which is what workload is inducing this? Where how it which is the question. I know, Adam, you and I have talked about this over the years that it's kind of galling that we, as software engineers, can answer the question of what is the draw of this workload? Like, how much power is this workload consuming?
Speaker 1:We, as a as a discipline, have not been able to answer that question. We have not been able to answer that question because of all of these sediment and layers that you can't cut across. And so you really do need you can't do that a little bit. You need to do the whole thing.
Speaker 2:And because no one can answer those questions, and a lot of people don't aren't aware of the fact that you've got, you know, up to 20% of power in data centers getting hoovered up into fan fan performance and and running the cooling systems in these in systems, which is galling. I mean, you should not have that kind of inefficiency for moving air in the data center. And that takes away from how much CPU you can get in there, DRAM storage, and, and and instrumentation is huge. But I think one other question in addition to what that person is asking, which is, why can't you do this? Is, should you do this?
Speaker 2:Because I think that's, you know, the the one of the questions that a lot of the companies we have talked to, have asked themselves is, like, why are we in the business of building a bespoke private cloud? Right? Our customers don't know we do it. They would be aghast if they realized how much of our team is responsible and and fully subscribed for dealing with server vendors and storage vendors and software licensing and integration and maintenance and upkeep and firmware updates and warranty dealing with warranty and part replacement. And
Speaker 1:And that's assuming everything works.
Speaker 2:That's assuming everything works. And, like, what if we could take that 500 person team and point them at doing things that our customers do care about? And the that's, you know, the that's the realization that a lot of folks are having once they have moved a certain set of their workloads to the public cloud is how do we get much, much, much more efficient on prem? And I I think it it becomes harder and harder to justify. I wanted to go build your own kind of bespoke on prem private cloud.
Speaker 4:It's interesting because, Brian, as you were describing it, even a single server is really this cacophony of different vendors, you know, all rolled up into a single piece into a single box. But then, you know, Steve, from your perspective, you've also got a bunch of vendors in the large software and hardware. So if there is some problem, get a bunch of vendors, none of whom really is is, incentivized to own up to the problem. There's a lot of finger pointing.
Speaker 1:Oh, absolutely. And this is true. Yeah. It's it's certainly true across devices or across a DC. It is true within the box, and, I I mean, Dell bluntly is infamous for for switching parts underneath the same part number.
Speaker 1:And, you know, I didn't even know that Toshiba made hard drives until they were in our data center, inducing a a performance problem. And it's like, I assume that we were running HDST. It's like, no. Nope. Nope.
Speaker 1:Now you're running Toshiba drives. And so the the inability to kind of control that. I mean, it's it's remarkable how, what I mean, what a cacophony it is. And and this is you know, you used Timothy Roscoe's term, Adam, is congealed. All of this congealed, and that's the problem.
Speaker 1:And it's gotta actually we actually wanna have it be designed, and that's what what we have endeavored to do.
Speaker 2:There's actually a great tweet, like, a year or 2 ago, where someone who had filed a ticket, and they had some issue with their Dell VMware install, which at the time, Dell owned VMware, and had said, perfect. I have an open ticket with VMware and Dell, and they have to work together on my support ticket. I feel like I'm dealing with my divorced parents. Exactly. And then just, like, expand that to, you know, 8 vendors instead of 2.
Speaker 1:Yeah. It's it's brutal. And, you know, there's a, Adam, there's a good question in the chat about, just in terms of that kind of rack scale efficiency. I'm not sure at what kinda what order we wanna take these things on. I'm not sure where this this
Speaker 4:the stigma has become, but, slots in given part of the goal of rack scale design is power efficiency through fans and backplane power bus bar. Can you tell us about the approximate percentage improvement in power efficiency over a rack full of commodity hardware?
Speaker 1:Yeah. And I think that and this kind of dovetails into another, I think, theme on those Hacker News questions, which is like, yeah. But, like, there's nothing left on the table. But you're talking about 1 tenth of 1%, was one of the comments. It's like, you're definitely not talking about numbers that small.
Speaker 1:And the reason is this, when you take a clean sheet of paper from the rack level, there are a whole bunch of things that you can go do that are actually not possible at the 1u2u, and part of it is indeed like the tyranny of that 1u enclosure. So when when every server vendor is trying is is incentivized to deliver you the maximum density in the smallest unit, they give you a 1 u or a 2 u that's kinda loaded with compute. The problem is that you actually don't even get to to a full rack before you've tapped your rack out from a power perspective. And what you've done is now you've got a rack that's like a runt of a rack. You've got a rack that's only like a quarter of the way up or a third the way up, and it is screaming.
Speaker 1:And I mean that acoustically, it is screaming because those small fans struggle to move air. So those small fans are operating at 15,000 RPM, 20,000 RPM more to move air. And as it turns out, like, fan, the air movement is like on the order of the cube of the diameter of the fan. And when you get to much larger fans, you can actually operate those fans at at with at much less rotational speed and move way more air. And then when you do that, you draw way less power.
Speaker 1:So in part of the way we get that that efficiency is we're not actually doing anything magical with respect to the CPUs. We are in fact, our goal is to run the CPU at its max TDP, at its thermal design point. What it's all of the rest of the draw in the system. So it is, for example, it is the our fans operate at 2 k RPM to or a plus minus. They tend to operate at, like, 25 100 RPM even when the when this the machine is cranked.
Speaker 1:And acoustically, by the way, that is relatively speaking silent. And it is striking when you see this in the the whole rack on in front of you, because it doesn't look like it's on. It doesn't sound like it's on because the the fans are so quiet, and they kinda go behind the rack, and you kinda feel all of this heat just dumping out on you. So it it is it is not about necessarily the the what we're trying to do from an efficiency perspective is make sure that that 15 k w rack budget, we want all of that to go into useful work as much as possible. We don't want that to be to to be wasted.
Speaker 1:So we want that to go to your compute. We want that to go to to your NVMe drives as you need it. We want it to go to your networking. We want it to to drive that to the things that you need to go do it. And the the the multiplier here can actually be pretty significant because, when we are because the the other thing that we've eliminated is we've eliminated the redundant AC power supplies that exist in that that those one u two u enclosures where you've got, the redundant AC power supplies that are converting from AC to DC, and no one running at scale does it that way.
Speaker 1:Everyone running at scale runs with a DC bus bar based system where you do your power conversion in one part of the rack, and then you run DC up and down a big honking piece of copper. And you can't buy that system from and the efficiency gains over there. Oh, the just just that gives you a really appreciable and important efficiency gain. But, the and and we're gonna have like
Speaker 4:Google and Facebook and and the hyperscalers all do it.
Speaker 1:Yeah. Alright. Go ahead. Sorry. I'm good.
Speaker 2:Hey. Just one thing, Adam. There's a couple people that in I saw were asking questions around. Gosh. It would be nice to see what this thing is like be but before having to write a big check.
Speaker 2:And, mentioned it in there, but we we are striving to have the availability of kind of lab infrastructure that, certainly, it's gonna be hard to fulfill a a a a bevy of requests, but that we wanna give people the opportunity to come in, kind of test drive, get familiar with, and look at the software experience of the system without having to, buy a full rack. So that's coming soon.
Speaker 1:And another really good question, an important question in the chat, and I sure. I'm more I wanna use this to tease an episode for next week. Asking about the I'm asking questions we love to hear about, like, hey. I wanna know about, like, shipping insurance, logistics, installation site, who does the install, like, let's get to write brass tax of this thing. Just operational questions around how this rack mechanically gets from where it's manufactured to me.
Speaker 1:And, Ariel, if you don't mind in the chat on that, I would love to let's take those questions, next week, because we're gonna have next week, we're gonna be joined by the operations team at Oxide, that has been absolutely just busting their butts for the last, you know, whatever, three and a half years, to get this thing shipped, and there's a whole lot we wanna go talk about. I wanna I want us to talk about the entering of the crate, because Adam, have you been to the the crate is itself it's its own, like, absolute ensuring marvel as far as I'm sure we're taking, like, 3 iterations, like, great.
Speaker 4:Yeah. Like, taking, taking I'm racking a rack or uncreating a rack rather without instructions was great. It was like it felt like one of these, kind of origami creations that
Speaker 1:was just
Speaker 4:intuitive and, like, the the tools were sort of there. It wasn't, like, hunting for a hex wrench or whatever. Yeah. It was awesome.
Speaker 1:So we wanna, I think we we wanna have an in-depth conversation on all that stuff, when we got the ops team because they're the they're the the ones you really wanna hear, and they are taking a, a much, much, much deserved, time off. So, we we wanna catch up with them next week.
Speaker 4:Should we go back to some Hacker News? Yeah. Absolutely. Go ahead. Alright.
Speaker 4:Sweet. So, let's see. One says, it seems that Okta is aiming to be the apple of enterprise hardware, which isn't too surprising given the background of people involved. Some used to be something like that as were other fully integrated providers, though granted some didn't write its own Unix from scratch. Almost like coming full circle from the days where hardware and software was all done in integrated fashion before Linux turned up and started to run on your toaster.
Speaker 1:Yes. Yeah. There's, and and that indeed hits a big thing. And we're certainly not, that we find things to emulate in both Apple and Sun for sure. I I think what we see that that both those companies have in common and by the way, have in common with a lot of other companies that people emulate is this idea of integrated hardware and software in a unified system.
Speaker 1:And, you know, one of the things that has been was certainly when we were initially raising very frustrating for us is how iconoclastic that idea felt, because it doesn't feel iconoclastic to us. It feels like, you know, integrated hardware and software. There's another company called Nvidia that's pretty good at integrated hardware and software. And and, obviously, Apple and, you know, Sun. I Adam, you and I would probably be a little more jaundiced about Sun back in the day.
Speaker 1:Sun definitely did it, but also didn't do it as well as we wanted them to do it. We really tried to be tried to implement that vision certainly at at Fishworks. And that's part of the reason why when we got to Eclipse, it was such a breath of fresh air because finally, we were talking, and I think we talked about this in our episode on hard tech investing that the it was such a a breath of fresh air to encounter an investor who's like who was completing our sentences. Like, you know, we've done the math and, you know, whatever it was. Steve, was it, like, 19 out of the largest 25 companies companies ever built have hardware and software together.
Speaker 1:So I I
Speaker 2:we It's not hardware and software together for for integration sake and for doing both. It's because that is required to be able to deliver what the customer wants.
Speaker 1:That's right.
Speaker 2:Which is this better experience, this services based experience, and you can't control that experience if you're only doing half of the equation.
Speaker 1:Right. You really gotta do you've gotta take this this holistic system approach, and that's been really important. So I think that that's what we would emulate in both companies, Adam. I think that the, and certainly, you know, we see that to to emulate, in Apple. I mean, we did have a, a VC at one point that was asking us when we're originally doing our raise, you know, what what what's the the best analog for oxide?
Speaker 1:And this is one of those moments where the mouth started moving before the brain was really in gear. And before I could really check myself, I had already said the a s 400, which is definitely, like and I you know, it's one of those things where I like I said, the a s 400, and you gotta get on the, you know, the rest of your brain to kinda doing the slow clap behind you, Adam. You're like, nice going. Nice going.
Speaker 2:Well a s 400.
Speaker 4:Side. If it had landed, you would have known that it was sort of it was meant to be. Yeah.
Speaker 2:If it had been that one right VC.
Speaker 1:That that's right. Well, it's
Speaker 4:true. I've been looking to invest in the AS 400 of 2020. Coming of the
Speaker 1:AS the AS 400 of the future is what I've been telling all my partners we need to find. And, of course, like, my brain is like, oh, yeah. Yeah. Don't say iPhone. Don't bother don't bother to say the iPhone.
Speaker 1:Ever heard of it? Never heard of the iPhone. Let's go the AS 400. Why not? Why not?
Speaker 1:Yeah. Give me a history lesson, Cantrell. You
Speaker 4:Everyone loves history.
Speaker 1:Everyone loves history. Exactly. And it's like, it's like, oh, yep. He's saying system 38. He's doing it.
Speaker 1:Well, there you go. This is you know, why do I bother? The but the Ace 400 was another example of this, of really delivering. And, honestly, Adam, what you and I did at Fishworks is an example of this, where what you are delivering is instant value for the customer. I mean, one of the thing, I mean, oh, Adam, I knew you had a have you had a chance to run Wigget yourself?
Speaker 4:Yeah.
Speaker 1:I was For sure. Oh, Mike, I have not actually run it. I've seen only other people running it, but we've got our own rack here that we run our that that we've been using to run infrastructure on, and we which has been great, obviously. That just says consuming your own product, you learn a lot about it. And, this week is my turn to actually run it.
Speaker 1:So I was doing the the fresh install on the rack today. I was giggling as I was running. I wake I mean, the the installation experience It's
Speaker 4:ridiculous. Yeah.
Speaker 1:It's ridiculous. It is so, I mean, it's it's kind of like unspeakably delightful. You know what I want? I want to have the first person who has suffered through Dell, HPE, Supermicro, Cisco, Arista, VMware, setting up a cloud, and then goes to wheels this rack in and connects to this thing over the technician port, and they are configuring the rack. Again, it's just this unbelievably gorgeous experience.
Speaker 1:Yeah. Totally integrated.
Speaker 4:Is that it's not a CLI. It is a textual based GUI. Right? It, like, takes over the screen, and it's just so gorgeous. And and and kind of defies my expectations of what one could do in ASCII.
Speaker 1:It absolutely defies expectations of what one could do in ASCII. And, you know, kudos to to Andrew and to Rain, to John, and to everyone who's been good. We're working on this together, but this is amazing. And, I mean, I was gonna I was giggling when I was doing it. It's so incredibly delightful.
Speaker 1:And what I am looking forward to is then the technician, the operator, the engineer, the SRE, that then goes to describe that experience to their peer who wasn't there, and it's gonna sound like they were having a fever train. It's just gonna sound like, are you listening to yourself? Like, this is you you just sound it's like, you know, you were in this Nirvana, this I this IT Nirvana where things magically happened. It's like the land of chocolate. But it was is really and the you know, that's the kind of experience that you and I was frank because I was just talking to Ray earlier today, and Rain's like, yeah, you know, I I really, you know, and her background is from Facebook, she's like, I really don't know what the kind of like, the what is the state of the art here?
Speaker 1:And it's like, very bad. Very bad.
Speaker 4:It's
Speaker 1:the state of the art is, you know, IPMI and writing down IP addresses in your notebook and then, you know, suffering through configuration screens, trying to hit f one at the right time and KVM
Speaker 4:and Wait. Like, trying to network boot one of these servers where you're logged in to you know, the two options are some h t, you know, HTML 5 versus Java Virtual Console, and you have to, like, figure out how to input the secret keys at the right moment. And, you know, often it works and sometimes it doesn't. It's it's lunacy. And and as soon as you have to go back to that, you realize how, how much disrespect maybe the industry has for the people running those commands.
Speaker 1:I I did. Totally.
Speaker 2:Absolutely. Putting it really, really well. Because I think we've we've we've seen, you know, some of the lower level systems access parts of the product that we've shown off to prospective customers are some of the things they've gotten most excited about. Because these are some of the areas of the systems they have to engage with that are completely overlooked and that haven't been because there's just no empathy for the systems operator, network operator, person who's gotta update firmware. And it it is I think that's it.
Speaker 2:It it it is just that there's just been no thoughtfulness that has gone into that, and it is, I'm I'm looking forward to test driving that.
Speaker 1:Oh, man. This is so exciting.
Speaker 4:That is ignorance, and some of it is, just technical obstacles. You know, like, the folks like, certainly, folks, like, at Google and AWS value the time of operators, at least in the sense of they want an operator to be able to manage as many systems as possible. Same thing that that other companies want. They just don't have access to the tools that would allow them to do that.
Speaker 1:Totally. That's a good point.
Speaker 4:Okay. So, maybe a tougher question here. I truly and honestly hope you succeed since, this Right. Right.
Speaker 1:Yeah. I'll hook it over. Alright. Alright. You know, we we embrace here.
Speaker 4:No buzz coming. No. I know for certain that the market for on prem will remain large for a certain sector for the foreseeable future. However Okay.
Speaker 1:Now yeah. Exactly. Now you're even more braced. However Right.
Speaker 4:The kind of customer who spends this type of money can be conservative. They already have to go with an unknown vendor and rely on unknown hardware, then they end up with a hypervisor virtually no one else in the same market segment uses. Would you say that KVM or ESXi would have been an easier sell or a harder sell here? Innovation budget can be a useful concept, but I'm afraid it's being stretched a lot. So the Yeah.
Speaker 4:Interesting question just in terms of, you know, how does this thing get consumed?
Speaker 1:Totally. And I think it was, you know, a bit of an open question. It wasn't open for very long, but we did kinda wonder what do we do about the hypervisor? Definitely, and and we we had some advisers to the company early who were like, you play. You really need to be sure you can run ESX on the EXI on this thing.
Speaker 1:It's, like, yeah. No. We're not gonna do that, in part because we know that we can deliver more value. We and we had a lot of experience, with our own hypervisor at at Joint, with SmartOS, Beehive, and kind of had a lot of experience in that domain. So, we we knew we could go do that.
Speaker 1:We also felt that there were a lot of reasons why VMware is not really popular with their own customers. And, you know, actually, that same early advisor to the company when, you know, VMware knows if it's gonna close or not, but when when Broadcom announced the acquisition of VMware, that same adviser was like, oh, you guys were so smart to not do ESXi. I'm like, this. Alright. But the, and and then began to rattle off a lot of supporting detail about how even in the last 3 years, I mean, just to be, fair to to their perspective, because, honestly, that perspective was has shifted in part because the company itself has shifted as it has really gone and become very rent seeking.
Speaker 1:So I think that, you know, the question of, like, hey. These folks are conservative. And, that's definitely that is true. I mean, I would say that, like, what what they're doing, I would I would phrase it slightly differently, in that what they're doing is important, and what they're delivering is important, and they treat it with with gravitas. They they they take their responsibility to their own customers seriously, and they need to know that they're getting a high quality product that a company stands behind.
Speaker 1:I think part of the frustration that they are currently feeling is, I'm frustrated with Dell. I'm frustrated with HP. I'm frustrated with Micro. I'm frustrated with Cisco. I'm also really frustrated with VMware.
Speaker 1:And then I have got and, by the way, like, it VMware being bought by Broadcom does nothing for my frustration. So I think that that comment is is correctly identifying some of of the the natural reticence to just, you know, deploy the latest technology or the latest thing simply because it's new, but it may be underestimating the level of frustration that people have, with VMware in particular, and especially as they're looking to go negotiate a new agreement with VMware. It's like they're they're not feeling necessarily great about it.
Speaker 2:Well and I think I mean Telecoms
Speaker 3:are airlines. Right? Like, you use the one that screwed you over least recently, and then you just move in a circle forever.
Speaker 1:That's right. Yeah. And, I
Speaker 2:mean, they're they're he's kinda keying in they are keying in on business risk. Like, you know, can obviously, new company. So can I trust that the company will be viable as long as I'm gonna be making an investment in in the technology? And this is a big part of being able to find the right partners, the right the the the right investment partners that were signed up to stand behind Oxide. And and, we we've been very, very fortunate to have a a a great team of investors around the company that have a a decade plus long vision for for what we're gonna go do and and have have been willing to jump on the phone with Fortune 500 companies to to talk to them about how they view the company and and the longevity.
Speaker 2:It's you know, we think it's super important that we are very transparent
Speaker 1:Yeah.
Speaker 2:About what we're building. Because, I mean, I would I would yeah. You can you can kinda flip it around, and you can say, you know, is it is it better to have a a something that someone else has run before that you know nothing about or to kinda know everything about the thing that you're now adopting? And what we've heard from customers is the fact that we are opening up the software stack, the fact that we've the fact that we've been as transparent as we've been about what we're building, how we're building it.
Speaker 1:Indeed, this very podcast, dear listener, is is a part of that transparency.
Speaker 2:And what we've heard from folks that are in this conservative set, right, this sort of fortune 1,000 demographic is it is precisely because of that that they've been able to socialize and build trust internally in the organizations that have allowed them to take the first step. And and I think you as you mentioned earlier, it's like, you know, that it was in question for a period of time, but not long. And, now we are we're shipping to that demographic, and it's it it also is gonna be very helpful as we get more and more installs there that it will make it more comfortable for the for the ones coming after.
Speaker 1:Absolutely. I mean, the proof is is always in the pudding. And, you know, I think that we and, indeed, you know, 3a half years ago, people were wondering about, like, boy, you're looking at AMD versus Intel. And making all of the same arguments in fact, we had another different adviser to the companies. Like, I think you're making a real mistake by going to AMD because no one's gonna adopt AMD.
Speaker 1:And we're like, yeah. But if you look at just look at the numbers. Look at our analysis. What what what hole are you finding in the analysis here? And, needless to say, we are not at all surprised by what has happened in the the the last three and a half years for the relative fates.
Speaker 1:And not that it, you know, never one should never discount, Intel's ability to compete, certainly, but, but we knew that, like, that AMD, by delivering a superior microprocessor product, would win people over, and you win people over with with a better product and better economics and so on. So we we believe that in the in in the limit, you know, we've got something right compelling here. And in the in the meantime, it is that transparency and trust that is so important to us. Yeah.
Speaker 2:And customers are looking for, you know, before they're asking, oh, what's the hypervisor? They're asking, like, what is the interface that my development teams are gonna operate against? It's like, do you will will Terraform work? Yeah.
Speaker 1:So
Speaker 2:and and they they, I mean, they care about the hypervisor in as much as they care in as much as they have upstack third party software constraints that they wanna validate. We have not run into those that are disqualified, and we're gonna continue, you know, being able to support more and more and more as we, as we go. But, they're they're looking for a bunch of things that include, like, when I have an issue with the application that I'm running on this infrastructure, how many different companies do I have to call?
Speaker 1:Right.
Speaker 4:And indeed, what Right now, it's, like, 5 or 6.
Speaker 1:The right now, state of the art is it's 5 or 6, and it is really asking for permission for them to tell you to go fuck yourself. I mean, because none of those 5 is gonna actually own the problem even when the problem is indisputably theirs. And this is the thing that is really, really, really frustrating is the cacophony inside the bomb, Adam. That's when you really get and we did it. We had a a brutal problem where we were having machines reset due to a dim failure, due to due to a dim error, and, you know, we get the runaround from the single vendor that that sold us the part, but this hold us the unit we're getting the runaround.
Speaker 1:And they're trying to blame they're trying to blame everybody else. They're trying to blame every one of their suppliers. They're trying to blame. And then, you know, we're the only customer that's seeing this. I mean, they're not ultimately and and and the kind of the realization was they are not taking responsibility for this problem, not because they don't want to at some level, but because they can't.
Speaker 1:They can't because they didn't actually build the whole thing. They actually don't know how it works. And that is a real problem, and one of the things that we believe that is our core differentiator as a company, because we do actually understand how the whole thing works top to bottom, we can absolutely take responsibility for for any problem within there. And as we, one of the things you're gonna hear us talk about is delivering oxide value. Delivering oxide value is the capacity to to support that for the customer.
Speaker 1:So when we when we look at a component, when we look at a part of the system, we wanna know that we can deliver oxide value with that component, and that which is to say, we wanna know that we can actually support any problem that that the system might have as a result of this component. Oh my god. As Zentranix says, usually, they blame the operating system first. Absolute oh, god. I mean, you you know what I mean?
Speaker 1:God. And and and, you know, I used to think because, you know, we were at Joynt. We were running this Alumos derivative smart OS. And that where I'm getting you you of course, the second they find out that we're not, like, wait a minute, you're not running Linux? Oh, well, then that's the problem.
Speaker 1:It's like, no, no, but wait a minute.
Speaker 4:Unsupported configuration. The end.
Speaker 1:Unsupported configuration. And so literally, we had a Dell perk issue, where the perk the machine would not boot because the perk had a parity error, and you're like, it's a smart OS issue. The perk is the is the raid controller. And they're like, it's a smart OS issue. It's like, smart OS is seeing through space and time and preventing this machine from booting.
Speaker 1:I I don't understand how, like, I You mean before
Speaker 4:we have even executed a Before we have executed an instructor.
Speaker 1:Yes. That's right. That's right.
Speaker 3:It's It's, like, reminds me when I was in college, you know, they didn't support Linux on the university network, but, of course, I was trying to do that. And so I'd, like, email, you know, tech desk help with, like, hey, you know, this isn't quite working. And they would be like, what does it say when you open this control panel and put that in? And I'd be like, oh, yeah. I definitely opened that control panel, and here's what it says.
Speaker 3:You know? It just, like, completely and then eventually, at some point, you know, they would, like, figure out I was actually using Linux.
Speaker 2:I'd be like, yeah.
Speaker 3:We just don't support that. I'm like, I
Speaker 2:bet you could I don't Well,
Speaker 1:and the thing that that I have, you know, since learned, because, obviously, like, fine. Like, your aunt, trust me, that we get really used to that answer to the point that, like, alright. Yeah. I get it. Fine.
Speaker 1:Yeah. It's our fault. Somehow we're preventing you from booting or whatever. The, but what you learn is that, like, actually, everyone gets that right. And it's like, oh, like, which version of Ubuntu are you running?
Speaker 1:Like, oh, run a different version of Ubuntu. Like, oh, are you running with all, like, all the patches? Like, download these patches. Like, this got nothing to do with my problem, though. Though.
Speaker 1:And and the and you realize that all of that stuff is actually an excuse for not understanding the problem, and it's a terrible customer experience. I don't care what you are a customer of, but to be told that I'm not gonna investigate your problem because of something that you're doing. It's basically I mean, it is quite literally victim blaming. It is like, you have a problem because it's you. I found the problem.
Speaker 1:You, dear customer, you are the problem. And you're like, you know what? Maybe I am the problem. Maybe I you know what? I have the problem.
Speaker 1:I'm gonna remove myself for this. And actually, on that perk issue, Adam, I was so frustrated. This is kinda like back in the day. This is in, 2010. And, being told in particular, the other thing they're telling you is, like, I it it had this is the the Dell reasoning why this has to be a smart OS problem.
Speaker 1:Because we're not seeing it from any other customer. And you're like, okay. Are you are you not seeing it from any other customer, or is no other customer telling me? It's, like, again, you may be speaking your own truth, but I I refuse to believe that we are somehow the only customer seeing this. And I gave a presentation, and there were like 300 people in the room.
Speaker 1:This is a surge 2010. I'm like, hey, just before I start my presentation, is anyone seeing parody errors on a Dell, Perk, whatever it was, h 700 or whatever it was? And, you see in a room of 300, you see maybe, like, 9 hands chewed up? Like, definitely not 0. And you could just see, like, the 9 hands all making eye contact with one another being like, wait a minute.
Speaker 1:Okay. I'm not the only one seeing this. And the, so with very much everyone had been told you're the only one seeing this problem, and it is a it it it's a terrible customer experience. And we wanna we believe that in order to be able to deliver better, you need to be able to deliver system, but then importantly, you can deliver better, and you can deliver a much better customer experience. And that's very important just to kinda to to kinda pull it back to that original question for those folks that are delivering very important workloads to their customers, that level of support is really important, and the knowledge that we can actually provide that is really, really important to these folks.
Speaker 1:And, I mean, the commenter's right in that these organizations are not they're they're they're not gonna buy the latest thing just because it's the latest thing. But they they also are, they've got a real problem in front of them, and they wanna find a way to solve it. And Oxide actually provides a vector to help solve that. Yeah.
Speaker 4:It it understands the the pain that they're in even though the pain is the status quo.
Speaker 1:Absolutely. Absolutely.
Speaker 4:Good good follow-up in chat. So what is the interface when I turn this computer on? What is the 0 to first value when I buy this hardware?
Speaker 2:Oh, great question. Great question.
Speaker 1:I mean, that is just yeah. That's the
Speaker 2:Especially given the time to value for for on prem today is, you know, we we've heard up to 90 days from the time all the boxes land in the data center to developers getting on and actually being productive. So what what the experience that we are delivering for customers is you roll the rack in, you have to give it power, and you have to give it networking.
Speaker 1:You have to check the height of your doors.
Speaker 2:You do have to check the height of your doors. But
Speaker 1:just the it it the straps are
Speaker 2:a little water bowl,
Speaker 3:and you give it a bed.
Speaker 1:That's right. And That's right.
Speaker 4:You don't want you don't wanna have the Let's work my puppy.
Speaker 2:Ideally, no speed bumps between the loading dock and the data center. Yes. But but you and and the experience that, Brian, you were talking about earlier, Wicket, the this kind of rack setup service that, but but you get it that you get it into the data center, get it to its, its floor tile, give it networking, give it power, and you are then off on starting the software experience.
Speaker 1:There's a modicum of configuration that you that it needs as a literal kind of bootstrap configuration. You provide that, and, yeah, then you're
Speaker 2:But the expectation is within, I would say I mean, I've been saying a day, and then you will scoff and
Speaker 1:say I will scoff.
Speaker 2:1 hour. I will scoff a day. Yeah. Yeah. Well, because you've been saying because a day is so much better than 90 days.
Speaker 2:I know. You've been saying, like, 8
Speaker 1:a day or a small number of days. I'm like, look. I know that normally you're making a claim, and I'm telling you to, like, cool your jets. But in this case, like, heat your jets up a little bit. Like, would you it's like no.
Speaker 1:It's like it could take a day. I mean, I guess, like, you know, maybe it it
Speaker 2:but but same day, you are opening up a large pool of infrastructure resources to your customers, namely software development teams, SRE teams, platform teams, and they are able to interface with an API, CLI, or a console to go create a project to programmatically, like, you know, script to deploy instances and configurations for the software they wanna go run, invite other teammates, be able to, integrate into the off system that they use today. So the experience should be, like, going to you know, for those that have run-in AWS or GCP, that you can go to an elastic compute service, storage service, and set of network and security services and be productive deploying software in a matter of minutes. And that should that that will all take place, well, I guess we're gonna say now, with within an hour or 2. Right?
Speaker 1:Yeah. There you go. Okay. I get it.
Speaker 2:Rolling the rack in the data Right. Which is, when when we have told folks this It's unbelievable. Yeah. They they've they've definitely been in disbelief.
Speaker 3:To to come to Steve's defense here slightly, Brian, I think your your last word there is totally accurate. Like, it's unbelievable. Right? And so sometimes you have to tell people an answer that's much longer than it actually is because they just literally will not believe you. I I used to work in the payment space and I did deal with banks, and one of my responsibilities was talking to, like, the vendors upstream.
Speaker 3:And they used to always laugh at our calls because, like, I would be like, I would like this tomorrow. And they were like, oh, we were hoping you'd say 2 weeks, but, like, the the person with the longest bet was, like, 3 months. And that's just because those time scales are, like, so much longer than
Speaker 2:they are in, like, our corner of the world that, like, you
Speaker 3:know, people won't believe you if you tell them it's only an hour when 90 days is their norm. So even telling them, like, couple days, like, seems hard to believe.
Speaker 1:It's hard to believe. No. It totally is. And that's actually yeah. It it seems like a total stretch,
Speaker 2:and it is so and so they can provision, you know, arbitrarily sized, instances, run arbitrary operating systems in those instances, and be able to kind of carve up the system to meet compute intensive, memory intensive, IO intensive, storage intensive, like, whatever whatever the particular application or set of instances require or software requires, be able to do that and do that programmatically. And so, again, for those that have used EC 2 and EBS and VPC like things, like, very familiar. For those that haven't, it it it's gonna be a a pretty remarkable experience.
Speaker 1:Well and, importantly and, Steve, you said in there, but I just wanna emphasize, you said doing this over console, over CLI, over API. And one of the things that I'm really proud of what the team has done is allow you to use any of those 3 vectors for more or less anything in the system. I mean, it is really remarkable. It's a huge credit to, I know, work that you and you personally have done, but the the, it's been and, god, the the the UI, the console UI, which I feel like maybe we have we have not emphasized enough certainly around here, is just it is so fast. It is really, really nice, and that team's done just a beautiful job, on the the the, the web console.
Speaker 1:But then also having those same interfaces available via the CLI, via the via the API allows you to pick whatever the right vector is. I gotta be told in the chat that you should talk about the UI more than the the 80 fans. Alright. Listen. I you know, we do not need to choose between the UI and the 80 fans.
Speaker 1:I can love
Speaker 3:them both. Both children equally.
Speaker 1:Exactly. You know, you know, when when when Bridge was pregnant with our 3rd, my mom had a great lunch. You find that that your your love expands, but your time does not. And my love expands. Alright?
Speaker 1:Old crow, my time ex my love expands. My time may not, and I will give more time to the UI because my love is I I I have love for for both the UI and the annual winner fans. Well, we Well, speak
Speaker 4:of time not expanding. Let's let's keep it moving. There's a But the 80 millimeter fans did come first. That's but, one of the concerns
Speaker 1:That was a good picture.
Speaker 4:One of my concerns, when buying a complete solution like an iPhone or an oxide rack, like, Key Face, is how much ownership I have over the device.
Speaker 1:Totally.
Speaker 4:Where does Oxide draw the line on ownership of the rack?
Speaker 1:Yeah. That's a great question. And and for us, it's actually really important that it's, it's your computer. So you you've bought the computer. It belongs to you.
Speaker 1:And one of our challenges is and this is a I mean, I would say this is without precedent, but we are shipping you a distributed system with that computer. So there's some special, like, challenges for, you know, how we support that, how we enable that, But the the the computer is yours. And It's very
Speaker 3:funny in the abstract to have to clarify that. Right? Like, it sounds it's like you purchased a thing. You have it. It's like Your money We actually do need to clarify that, but it's just like in the abstract.
Speaker 3:It sounds
Speaker 1:totally ridiculous. It is it is true. Yeah, Steven. You know, I don't know about that. But, like, yeah, this is where we live now where we actually need to clarify
Speaker 2:that, like, we are The thing you bought.
Speaker 1:The thing you bought is actually not a lost leader so we can sell you more ads. This is actually, like, a thing you bought. The end. So in terms of, like, the the the ownership of the system, it is actually it it is important to us. And it's important to us as an abstract point, I think, and important to us, individually as as technologists and and some place of outside that it is your computer, and we want to really enable that that ownership.
Speaker 1:And I think that that extends in a lot of different dimensions.
Speaker 2:And we wanna be transparent about so you've got a operator level and a developer level, kind of an end user, customer of the system for our customers. And, much like them setting up accounts in the public cloud, they will be setting up accounts and and provisioning infrastructure and operating at that later. And you'll have folks in many organizations that are responsible for kind of maintaining and managing the infrastructure that software teams are running on at what would be, you know, an SRE or platform team kinda layer. And then you've got the operations team, traditionally, that are responsible for, kind of care and and upkeep management of the lower level systems infrastructure. And so we will have a a a set of operator interfaces that allow them to understand what the system is doing and make sure that they can kind of manage and maintain and set expectations internally.
Speaker 2:And then anything that we are kind of injecting ourselves into by way of kind of support and updates, etcetera, is to, Brian, your point, just making sure that this complicated distributed system is running well. It's it's in it's in the interest of making sure that the customer is getting, you know, a highly available and efficient system that is running. So if there are places where we are, where where customers trying to go into the lowest layers of the system could put that at risk, you know, the the you know, we're gonna be upfront and transparent around that.
Speaker 1:So in one of the just to sharpen that point, Ian is asking, do we sell support contracts? And does it come with a certain number of years of
Speaker 2:of updates? So we do. It it comes with a subscription that allows one to get all software updates and and, again, all hardware warrantied. And, those come typically in 3 years, but, we have folks that are going to be getting them longer than that and shorter. So there are some folks that can do it only annually, and that is fine.
Speaker 2:We're we're very flexible on the on the timing of that.
Speaker 1:Another great question just to follow-up to that is the is the rack dependent on any oxide cloud services, or is it entirely standalone? I don't mean to laugh because I just I just laugh at Josh's suffering every time I think of physical infrastructure that relies on I we got a bunch of ubiquity gear that was maybe a debatable decision. What's that?
Speaker 4:Mistake.
Speaker 1:Yeah. And, the, that thing has an AWS console that apparently, like, I does like, works 9 to 5. That that console. That console is not extremely frustrating. So this is a long way of saying, emphatically not.
Speaker 1:In fact, it's very important to us that the oxide rack is entirely stand alone, that it is not going to oxide for any cloud services. So Well, I very, very important.
Speaker 4:However, for updates, it will fetch it from what is arguably a cloud service, that is to say, like, a web server running somewhere. And, you know, there
Speaker 1:are Yeah. And yeah. That's on yeah.
Speaker 4:But but but, I mean, your your broader point is is the the most important one, which is, like, this is not like AWS outpost. This is not some, you know, vassals, some some piece of a larger cloud. This is your cloud running independently.
Speaker 1:I just think they should rename an outpost to vassals. Like, oh, we're I'm
Speaker 4:using outpost has kind
Speaker 1:of that
Speaker 4:that sort of kind of
Speaker 1:thing. It does have the punctuation. Like, let's go all the way. Let's just call them bastiles. Totally, Adam.
Speaker 1:That's a very good point that, like, ultimately, in order to get updates in and then the other thing is, like, okay. What what if I wanna get support data out? Those be and that's ultimately, but the the the the customer has total agency over that data in and over that data out. That that's not happening, with the kind of an implicit cloud service. There was another yeah.
Speaker 1:Yeah. Go ahead.
Speaker 4:Please go ahead, Adam.
Speaker 1:Go ahead, Adam. Yeah. Yeah.
Speaker 4:Well, I I would even change it back to some of the comments in Hacker News, just because we've been, there there were some acerbic ones, but I did wanna read a couple of the nice ones. Someone says, this is super cool. I realize a lot of HN folks might not see the point of this, but it literally saves an entire team of people for companies. Another one says, oxide is such an ambitious project. I'm such a fan of the aesthetic and design and, of course, transparency of all the people who work there.
Speaker 4:I'd love to have a rack or 2 someday. Call us. And and just, not to cut
Speaker 2:you off thing, mentioning that, the the good comments, but just to reinforce one of those points, there's a demographic of folks, the the large cloud SaaS companies that, we don't talk a lot about, but we've had conversations with some of the the kind of the biggest, most well known, and their ambitions don't lie in, you know, this this online conversation about cloud repatriation and cutting costs. I mean, that is of interest to these these these firms, but they're more interested in, like, how do we grow our business over the next 10 years, and how do we capture more data that sits outside of the public cloud? And for those companies, they're thinking about how do we extend our infrastructure beyond just running on top of 1 of the big three cloud providers. And I can tell you, they there is there is no path that includes, let me go compile, you know, 6 or 7 vendors worth of infrastructure and build that team, Adam, that you just mentioned, which is, like, saving someone building a huge an entire team. And so I think that's just, again, kinda speaks to, yes, there's there's a a a need for operational efficiency in classic on prem enterprises.
Speaker 2:The cloud SaaS folks are trying to figure that out too because they can't go build, this this antiquated team or this team for legacy solutions to go extend beyond the public cloud.
Speaker 4:So I have one comment Yeah. That I had to read just for me because I had to read through all these Hacker News comments, which you know is not my favorite. This person writes, FYI for language pedants like me, it's on premises or on prem. A premise is something assumed to be true. So I thought it's a GSA that I endorse.
Speaker 1:Hey, Adam. I can only assume that this is one of your alts. That is
Speaker 4:Yeah. Whoever wrote that, what a genius.
Speaker 1:What a what a what a genius. And I and I mean, clearly, this is, if we put your supercut in the oxide channel yet, we definitely need I know we talked about the supercut before. The supercuts could you describe the supercut again? Because it is just I mean
Speaker 4:Well, so I was, yeah, I was watching re Invent, and, I I was there, but someone told me whenever you go to re Invent, you just watch the keynote from your hotel room. So I was following that sage wisdom. And, Andy Jassy got on stage with Pat Gelsinger, then the the, CEO of VMware. And Andy Jassy kept on talking about Outposts on premises, doing work on premises, having VMware on premises. And then Pat Gelsinger comes out and starts saying on premise infrastructure, on premise compute.
Speaker 4:And it was almost as though Jassy was overemphasizing it to try to correct him and pull him back.
Speaker 1:Are we talking about on premises infrastructure? Yes. I'm talking about on premise infrastructure. Like, we are in an impact then, friends.
Speaker 4:Oh, I see what you're saying. The premise is that the infrastructure is in your data center.
Speaker 1:So you you did a supercut.
Speaker 4:Yes. You yes. It took a very long time. I'm I'm not very good at video editing. I'm only slightly better at audio editing.
Speaker 1:But it was, it really is delightful, the supercut. I would like to believe that supercut has paid dividends because it is really very delightful, that Gelsinger and and jassy, of course. Jassy being correct in this case that, it is it is on premises. Please. Adam, my question came up in the chat that has also come up a bunch.
Speaker 1:It did come up in the Hacker News threads, has come up in the past, and is is, one that I, that I definitely wanna tackle is a reasonable question is like, Hey, how about something smaller? Like when is oxide gonna make a, a smaller rack? Like I, I, I like this, but I, I want it for my home lab Or I like this, but I want it for, the community college where I'm an IT administrator could really use this. And, I just wanna speak to that a little bit because, we definitely, well, the home lab probably like, home lab sorry. Love you, home lab.
Speaker 1:Probably not gonna be for the home lab. You could like, come on. You well, you you love to tinker. That's part of the home lab. And, for the home lab, good news.
Speaker 1:It's all open source, and should go play around with like a nuclear board and and, there's a bunch of stuff you can go do in the home lab. For that though, that kind of small enterprise user, we definitely see you, and we, our belief, and I think remains that we wanted to tackle the whole rack problem because this is a problem that is easier to scale down than up. And I think one of our concerns when we started the company is if you start small, it is, it can be really, really hard to get to the size that you actually need to get to for these large multi rack enterprise customers. So, we are focused on those large multi rack enterprise customers right now. But we have done so with a deliberate eye towards a an architecture, because we absolutely see those small to medium sized businesses, the car dealerships, the community colleges, the retail establishments, the the the retailers, the the amusement parks, the attractions.
Speaker 1:Like, we we see those use cases, and we know that, like, a 9 foot rack drawing 15 k w is not necessarily the right fit for all of those, but we have done we have built this rack with the end, which is not to say, you know, yeah, I would love to tell you we've got something immediately forthcoming. We definitely don't, but we've also, architecturally, there is nothing here that won't scale down to a point. It's not gonna go ultimately, it is not gonna go to the home lab. Ultimately, it is the the there there are limits to to where going under the desk. It's not going under the desk.
Speaker 1:We definitely a bunch of us do have gimlets at home. Definitely. But but I think part of
Speaker 2:the the questions have come in in a couple of different dimensions. One has been, hey. Can you scale this down from a power perspective? And the answer to that is yes. Yeah.
Speaker 2:Right. So we we've got folks that are interested in the the 9 feet is not a constraint, and they they have room for that, but not for power. And so you actually can deploy it as is with fewer sleds, for sure, which would bring that power requirement down quite a bit. The other thing that's great is that as we think about a future where there's different form factors, you know, a big part of this was getting the software right first. That's right.
Speaker 2:And so not having to go build a different set of software to go support a smaller footprint, I think we've done we've done a lot of the heavy lifting to be able to support those form factors. We just have to figure out what the broadest ask for the market is gonna be once we're out of or or beyond the core data center or rack size kind of edge deployment use case.
Speaker 1:And I would say we this is one where we're definitely keeping our ear to the market. So if you've got a particular use case where it's like, hey. You should hear about my use case, we would love to hear from you because we really this is one of those that we're gonna wanna be, really, paying close attention to because, it it is something that is just it it requires so little from us architecturally. It is, and then I think the other one that we probably wanna tackle, and sorry, I know, Tom is here. It's great to you it's great to see you, Tom, in the chat.
Speaker 1:And Tom's like, god. I'm so relieved that no one has said AI or Edge. So, of course, Tom, by just saying that you've summoned me saying, at least the those AI workloads. Because another question that comes actually, surprisingly, I don't think did come up in the Hacker News thread, which is kind of interesting.
Speaker 4:Didn't. But, yeah, didn't people didn't seem to latch on to GPU or or in particular or lack of support for GPU.
Speaker 1:Which is kinda surprising. But, actually, maybe that just reflects that we, you know, we are, people actually understand the need for general purpose compute in addition to I will obviously, you know, that that high performance compute on the g p gpu, we obviously understand the need for. I talked earlier about delivering oxide value. It is hard for us to see how we deliver oxide value with NVIDIA. NVIDIA is a very proprietary company, and one of the things that's very important to us is that end customer has a problem, we need to be able to support them all the way through.
Speaker 1:And with Nvidia, when you've got a problem, it's really hard, when when your version of open source software is a is a trampoline into your proprietary firmware, really, really hard to actually go support that end to end. So it's it I don't think it's impossible to see how we do with NVIDIA, but it's really hard. It's a strain, and NVIDIA is not exactly I think, NVIDIA is, I I I don't necessarily know that they they care about Octide. So, but I don't care about our opinion on that one anyway. So as we are looking to those workloads, again, wanna move very, very carefully, really listening to the market, but we are, definitely see some interesting things out there.
Speaker 1:So, we're paying attention to some of those things that are not necessarily NVIDIA. We obviously know that kudo barrier is a huge one. Again, NVIDIA has you know, they're very proprietary, but they have properly invested in hardware and software together, and it really shows in the depth of their moat. So, it's someone in the in the chat is talking about AMD GPUs. Obviously, we've got a very close relationship with AMD.
Speaker 1:We love what they're doing there. That's something to pay really close attention to. And just in the abstract, we are paying attention to the space, but we, that's not something. Right now, our product is really focused on that that compute storage networking.
Speaker 4:I mean, like, our product, if you compare to AWS, is, like, 4 or 5 of their services at best, and their service menu stretches across many, many screens. For any given sector, there's 5 different services they could bring to bear, so there's lots of opportunity for us to fill out that road map.
Speaker 1:That's right. Another good question was around sorry, Adam. Go ahead.
Speaker 4:No, please.
Speaker 1:I was gonna with a question about Firecracker, the and, is there a could you compare Firecracker to your hypervisor? We actually looked at the hyper Firecracker. We took a pretty deep look at Firecracker. The there are things we like about it. The it's, it's in Rust, which is great.
Speaker 1:It's kind of a de novo notation, which is nice. I ultimately concluded that it didn't it just didn't have the same use case. We are really focused on that, for lack of a better word, industrial VM use case where, Dan's pointing out the chat, the Firecracker is based on cross VM. We are really more focused on it's very important to us that we can run Windows as a guest, that we can run arbitrary Linux guests, that we can that we can run, yes, SCO open server as as a guest. I know if Super important.
Speaker 1:Super important. We, where if if Josh Blue Her needs
Speaker 3:to run or else
Speaker 1:somebody would buy it. Well, and I this is actually I I'm really just I I am actually trolling one of our number, the, years ago, we had a, a customer who's emailing, emailing us back in joy and days and really needed SCO to be running. So but we we it's important to us that we can run effectively arbitrary guests. So and Firecracker just doesn't have that same ambition, which is fine. The other thing that's really important for us, and, you know, Jordan, I saw mentioned it in the chat, something that we have not spoken a huge amount about, we definitely need to have an an episode dedicated to it at some point, is live migration.
Speaker 1:So another thing that we saw, and this is this kind of comes from some hard wisdom that we had learned, at Joyant that the the if you don't allow for transparent workload migration, And you you Adam, you remember when when VMware did VMotion. I think there was a certain degree we were like, really? That seems like is that a is that a toy? Is that real? And I don't know.
Speaker 1:Maybe, maybe, Adam, you didn't have that same
Speaker 4:I mean, I think there was both. I didn't understand the use case, and it felt like black magic. Like, terrible, awesome, amazing.
Speaker 1:And but turns out v motion is really, really important. And the and the ability to live migrate a workload is really important because if you can't do that, you end up with these islands of compute that you can't do anything with. And it so when people talk about the the inability to maximize the utilization of a DC or of a rack, it is often because it's like, no. No. I can't go per and, yeah, I know that that machine looks like it's 80% available.
Speaker 1:It's actually not. I've got one workload on there that I'm not allowed to to turn off, and I can't migrate it, and that machine is, like, out of warranty. So as soon as that workload is turned off, like, that thing's being deracked. So the and we had this problem a lot where we it was very hard to do life cycle management if you can't live migrate. So one of the things that was important to us as we're looking at the hypervisor and the use of a machine model was the ability to build live migration, which the, the team has done just an extraordinary job of doing.
Speaker 1:And that that, a a demo for the ages, was Greg and James in a shared Minecraft server that was being furiously migrated around the rack, which is which is great.
Speaker 4:That's awesome.
Speaker 1:But so the that that was another, very, important, dimension as we were looking at a hypervisor is the ability to go to exactly that, and another reason why Firecracker was not a great fit for us.
Speaker 4:There are a couple of simple questions on how can we do this? Some of which I think have simple answers, some of which have complicated answers. One is, are you shipping Milan racks? And if so, what does moving to Genoa entail for oxide? I'd also note that there was another comment, which I won't bother reading, that said that listed it as a Read it.
Speaker 1:Yeah. Okay. Wait.
Speaker 4:The listed items are for last gen CPUs and 2 gen old NICs. Go with the
Speaker 2:rest of
Speaker 4:it too. Also ZFS rather than SEF. It's hard blah blah blah. Anyway
Speaker 2:Yeah. There you go.
Speaker 1:Yeah. I mean, I do think it's kind of, like, ridiculous to be like, these are old generation CPUs. It's like Genoa was literally I mean, the release of Genoa was in November of last year. If we were on Genoa, we would not have a product for another 18 months. That is the reality.
Speaker 1:So,
Speaker 2:no one's using 2 by 100 nicks anymore.
Speaker 1:No one's using 2 by 100.
Speaker 2:Exactly. Meanwhile, there's no, like, I don't think I could I could I could saturate 10% of that. Right.
Speaker 1:I know. Oh, god. Just ridiculous, but the we are based on on Milan. So 7713 p single socket. Great part.
Speaker 1:Genoa, obviously, we have been Andy's been a terrific partner, and we absolutely will have a Genoa based lead. There are some challenges with Genoa. The, Genoa has got the the GDP is higher, in a way that's pretty challenging, and we are gonna wanna be take a a a again, keep a very close ear to the market in terms of what the market needs and for compute density. I mean, we know that from a compute density perspective, the 7713p crushes it. It's a great part, if you look at at the course per watt.
Speaker 1:It's really terrific. And I think as we're looking at the general SKU stack, we're figuring out exactly where we're gonna tack in there. And I you know, we we spend so much time on the MB road map. I don't I can't remember what's NDA and what's not, so I'm just not gonna speak to it, but we're also looking at at what comes next from Genoa. There's a bunch of stuff interesting there.
Speaker 1:We are certainly going we will do a a design around the their next gen socket for sure. So that's I mean, the the the big lift is the next gen socket, and we are absolutely, you will see a a general based product from us, but that is And and, importantly, architecturally, the rack level design
Speaker 2:is such that you deploy a rack, a set of racks in your data center. You then have the opportunity to go add sleds that are the next generation sled in that same rack or set of racks, kind of mix and match so you have, you know, modularity in terms of what you need. And it's not gonna require ripping out an oxide rack, putting in a new one.
Speaker 4:Because the rack itself was designed for you need to sustain several generations. Is that right?
Speaker 1:Yep. Yeah. Absolutely. And that we I mean, we view the rack as being really and that's the platform under which we're gonna go build. And so people are asking about the the actual, throughput of the rack.
Speaker 1:So, yes, it it is 2 by 100 dropped in to to each sled. So the, we we've actually we've got, 6.4 terabits of of networking, that faces the rack, and then another 32 ports on the other side that faces the network. So that faces the, customer's network. And Arin, made a a terrific observation kinda early on that, like, hey. Wait a minute.
Speaker 1:We can actually just use some of these front linking ports to directly connect racks together and actually be able to, grow to some number of racks, without having to actually add additional core networking at all, which is really kinda neat. So, we'd we've got, I I think, ample networking. I would also direct folks to the, our our episode on a a measurement but a measurement 2 years in the making, where we talked about, all of the challenges, of actually, getting to, to 28 gig PAM 20 gig NRC, and then trying to get to 56 gig PAM or, and all of the kind of the SI challenges there. But, I think we've got ample, despite your hacker news comment, Adam. I believe that we have got ample throughput on our cable backplane and our switch.
Speaker 4:Another question. Do you have liquid cooling solution collaborators?
Speaker 2:So our first, first collaborator in the kind of augmented cooling space is, with a company that designs, chiller doors, a company called Motive Air, and, we do not do liquid cooling in the system. So the our our first product, does not have liquid cooling, as part of the architecture. But we where we have, where we are deploying, where folks have, have additional considerations where they need to pull hot air out, we have worked with, companies that are building chiller doors to be able to capture that air and then convert that to room temperature air. And we're we're looking at, you know, what makes sense in future architectures, but not not today.
Speaker 3:So I think controlling the RGB lights that are all over the rack. I think that's the real question people wanna know the answer to.
Speaker 4:Exactly. The,
Speaker 1:go ahead, buddy. Well, and I I I think that, you know, one of the, I think the the challenge that we see in front of us more directly than the liquid cooling is just the power budget for the rack. I mean, one of the big, you know, and this is actually a bit of a misunderstanding too in that that Hacker News thread about kinda what our target is. We are targeting the enterprise DC, not the hyperscaler DC. And what is the difference there?
Speaker 1:The difference there is that in the enterprise DC, kinda lucky to get to 15 k w per rack, which is where we we are targeted at 15 k w per rack. There are plenty of enterprise DCs that are at at 10 k w, 12 k w, or less, you know, 8 k w that are already even less, versus you go to the hyperscalers, and they're like, no. No. No. Like, we are at at 35 k w or 40 k w or 40 like, we are I think the big challenge on as we begin to look at alternate cooling, the first problem we're actually gonna have is power budget in the DC, and I I power budget per rack, and kinda how we spend that.
Speaker 1:So that as we go to look, even and even going to partner MotivAir and and chore doors, with the and as we go to look at this, we're gonna look at it systemically and make sure that we're not you don't wanna have a liquid cooled solution that can only be a quarter of a rack. That doesn't actually solve any problem.
Speaker 4:This is a maybe a little bit of a can of worms, but I think important to address. Why do you use Illumos?
Speaker 1:Yeah. Sure. Yeah. That's a great question, and and that came up, in in the chat as well. And there are, you know, there are a bunch of different answers to that.
Speaker 1:We we very deliberately did not I think people would think that it's like, well, of course, you're using a link. Like, you're a bunch of joint people and ex Sun people. Like, you know, what else do you know? It's, like, well, no. No.
Speaker 1:No. Actually, we were very deliberate about that decision. We wanted to be very deliberate about that decision, and really wanted to kinda look at things afresh. And, you know, one of the things that on and and there are a bunch of different angles for this. The the I mean, the answer is there are a bunch of different reasons.
Speaker 1:Not least, we need to be able to control our fate, and this is a foundation that allows us to completely control our fate. Yes. There are features that we we need. Yes. One of the features is, Adam, the one that you and I implemented, and, I mean and it is true that, like look.
Speaker 1:I mean, we spent, a really significant portion of our career solving a really important problem, and it would be gutting to actually lose that solution. And, Adam, you know, you and I have been talking about this as we get more and more into production, but the work that you and Ben Nacker did to support USGT probes in Rust, which we knew at the at the time, like, this is important. And the yeah. If you're asking, wait, wait, what did you turkeys do? Was we developed this thing called DTrace, which allows you to integrate instrument the system.
Speaker 1:And it's been, it's been ported to a couple other systems. It's actually been ported to Linux, but it hasn't really been, it it but, has not been accepted to say the least. Linux has got, Linux has got, like, 4 different solutions, each of which does kind of somewhere between 60 to 80% of of what Detroit's does. And, we you know, the ability again, Adam, you and Ben added that kinda early. And I remember Patrick who's working on propolis is like, when is that thing done?
Speaker 1:Because I need it, like, yesterday. But, boy, as and we, again, we knew this. It's not a surprise. But as we have rolled that out, I mean, looking at Allen and Crucible, which is our storage subsystem, and the use that we I mean, we are using DTrace all the time on this thing in production to understand what it's doing. And the hell if I'm gonna live without that.
Speaker 1:The hell if I'm gonna live without that. And then now even said, I, like, I I don't think I've ever gone into with a more of an open mind into BPF trace and eBPF as I did when we were kinda doing that process. And, it's a mess. It just doesn't solve the problem we need to solve. Like, I you know, I I'm not trying to disparage it, but, the reality is it doesn't actually solve the problem we need to solve.
Speaker 1:It's got a totally different disposition. It's not focused on production systems. And I need the ability to support a customer at at our shared darkest hour, and I need the stuff that we have built to do exactly that. So the I mean, that's kind of maybe a personal reason, but there are were a bunch of other reasons. It wasn't just that.
Speaker 1:It wasn't just CFS. And, you know, one of the one of our colleagues, Laura Abbott, who I know is gonna be hearing the chat, and and, you know, Laura had a real Laura, formerly at Red Hat, had a really important observation that like, hey, just so you know, like, if you're signing up to do Linux, you're actually signing up to do a distro. And I I kind of, like, intellectually understood that, but over the over time, I I have come to appreciate that much more viscerally, where the because Linux ultimately is an operating system kernel. It is not it is not
Speaker 2:I'm waiting
Speaker 3:for it. I was waiting for you to say it was that Simpsons meme where it's like say the line, Bart, Say the line.
Speaker 4:Say the line. Say the line.
Speaker 1:Say the line. If I may, I'd like to interject. The, and so it is, you have to go make not one decision, but all these other decisions because things like the debugger is not they're not integrated. Libc's not integrated. You gotta make a libc decision.
Speaker 1:And now you're having a now you're a glibc maintainer. And the the, you know, kind of as time went on, I kinda and looking back on it, I'm like, that that, you know, Laura is really, I think that's kinda coming from her own kiln of pain and those, those who have come from AWS and so on have pointed out, like, no, it is the it is really, really tough. It's burdensome on a team to go do. And with we have so much more in the system. And, yeah, I mean, obviously, we are responsible for all the components, and, yes, you know, we've got a there's there is an element to which we are always gonna have to figure out where we are with respect to our robust version and our open SSL version and so on.
Speaker 1:So, you know, we we yes. We have a certain amount of that we have to sign up for, but what we don't have to sign up for is picking which LibC we maintain. And we ultimately because it's a community we know well and we're able to participate in, you know, Jordan made this point earlier that everything we do, we upstream, and that's really important to us to be able to not diverge from what the community is doing. And, you know, you may have heard me say this over the years, but, the, you know, we really believe in that. There's a great power to small communities.
Speaker 1:People in small communities always assume that the larger communities are are in but large communities, and, Steve, I know you know this very viscerally, large communities have, have different challenges. And, you know, not really concerned about, you know, in a Lumos Foundation feud or, you know, the I just there are certain things that we're just don't have to worry about because it is a small community, which we have, like, a terrific overlap in values. We also knew, I think, just to and and maybe I'm maybe we could maybe we're gonna post the already dead. He's, he's already dead from the Simpsons, but the, I also think that, and I talked about this when we talked about Holistic Boot, we knew we were gonna have a very, very, very different model for booting the system. And the the booting the system, this holistic boot and bringing up this kinda lowest layer of platform enablement was gonna require some really serious surgery, and you wanna have a, a community that's gonna be receptive to that surgery.
Speaker 1:And, you know, one of the things we hear over and over again is folks end up doing a bunch of important Linux work and then can't get it back. And, indeed, we saw that. Right? We saw that with Chris at Oracle, Adam, and Detrace, where Oracle actually ported Detrace to Linux. And it is done, and it worked, and it was great, and the Linux community had zero interest in it.
Speaker 1:And this is, like, a decade ago. Right?
Speaker 3:Even just now, like, you know, y'all put Rust in the Lumos kernel or deeper down in the stack before Linux get their support, which is still pretty nascent. Right? So, like, also with the amount of Rust we're using, the stuff we're kinda doing like that, that's an example of a move that we could make that we couldn't have if we want with Linux.
Speaker 1:Or at least
Speaker 3:it'd be much harder at the whim of someone else, like, at the whim of upstream.
Speaker 1:Yeah. And I and, you you know, I I think that the, and, you know, I think that and and the point that's being raised to chat is that the, and, honestly, actually, it's funny because Jason King is is raising this point about the the kind of the sabotaging of CFS in Linux, which actually, Jason, what was funny is that was happening exactly as we're doing the deliberation. And the funny thing is it is happening at this moment that my mind had never been more open to running Linux. And as that's happening, Torvalds is issuing these absolute screeds against ZFS that are not well founded. And I'm like, yeah.
Speaker 1:Okay. This is thank you. This is a, and then I think, ultimately, the kicker, Adam, was was, me talking to one of our former colleagues who had gone through a replatforming exercise. Because the I think one of the challenges is it's actually hard to find people in the Linux community who have had a lot of experience with DTrace. It just you you just don't tend to find it, that people who've got a lot of experience with DTrace in particular tend to stick with platforms that have it.
Speaker 1:There's a there's a important truth in there. And so, this is I don't think I'm betraying his confidence to say I was talking to George Wilson, who's a former colleague of ours, Adam, and what the one that you and I behold in highest regard. And asking George about this replatforming because I just gotten. I'm like, but don't you miss dtrace? He just says, every day, man.
Speaker 1:Every day. And and I'm like, okay. And and the thing is, like, George, on unlike, I think perhaps some people, George was he needed eBPF to work. So he was, like, going deep on eBPF, and he's and you going at length about the challenges he was having. So it was not like, oh, he just this person is just, like, more comfortable with DTrace and doesn't wanna use It's like, no.
Speaker 1:Wrong. This person actually has a customer problem to solve, and he will do absolutely anything that needs to be done to solve it. And he's being crushed by the inability to solve it. So that's probably too long an answer to that question.
Speaker 4:Well, you you raised another great point in there of the you know, we had a problem to solve, and our problem was not, you know, an OS problem, but rather, how do we deliver this platform that can provision virtual machines. And the operating system we choose I mean, the customers are welcome to ask that question. But if they need to know the answer, we've done something wrong. They should be insulated from that decision. And with the time line that we had, I think, unequivocally, we chose the thing that would let us go fastest.
Speaker 4:It might even it might help other teams, but certainly for the team that we had and the team that we could build, it was not even close. We would not be where we are today.
Speaker 1:That is actually a very important point, Adam. I think is that and I think that and, Steve, I think this goes to, like I know you had a reply on that thread about, like, hey. This is, like, this is something that the team was really familiar with, and that this is where the familiarity is not, like, the, like, a a fear of change or fear of something different. It's the no. Like, this is what we're actually fast with.
Speaker 1:And I think you're Adam, you're absolutely right that we would be, it would it's just I as with many of the decisions we've made, the alternative is too terrible to contemplate. Yeah.
Speaker 4:I mean, look. It's something we are biased. Maybe the even if there was some bias, it's also like, well, do we take what we know and the experience of the team and write that down to 0? It's like, no. We wanna leverage that as best we can.
Speaker 1:Yeah. And there's also just another question. I'm I'm really I'm sorry to do this. It's like, here we have arrived at open source licensing. How do we get to open source that's got nothing to do with oxide, and that is this kind of argument that is made by the Linux community about ZFS because of mistrust of Oracle.
Speaker 1:And to be clear, like, you are under no one is claiming that anyone is violating the CDDL by bundling ZFS with Linux. The only complaint would be the only question would be around the GPL compatibility. So you're actually there Emily in the chat is saying, like, I think if anyone had a mistrust of Oracle, it would be me. And it would. I was like, absolutely.
Speaker 1:Like, I know exactly I have got zero trust of Oracle. Again, it's like it's in part because I to trust Oracle or even to mistrust Oracle is to anthropomorphize Larry Ellison. So it's like, I actually I I I do I do I mistrust the the the lawnmower? Do I trust the it's like, no. It's a lawnmower.
Speaker 1:And, in particular there you go.
Speaker 4:Now I
Speaker 1:really said the thing. Yeah. You can really post it. But, in particular, that there is there is in no way does Oracle have in fact, what what Oracle had, would the real question would be, around patent enforcement, and this is the reason that it is not under the GPL is because we wanted a license. One of the reasons, one of the very important reasons is we wanted a license that was very explicit with respect to patents.
Speaker 1:And if you are a user of CFS, you have a license to the patents on CFS. And if you are a user of DTrace, you have a license to those patents. It's extremely important. So, yeah. Thanks, Adam.
Speaker 1:Sorry. Or just, like, it could just be blank space too. You can just, like, actually, like you don't really listen to that thing. I swear or he I you know, I'm looking at the chat. He said what line?
Speaker 1:Let's see.
Speaker 3:What were we talking about? Why is there a drop of blood dripping from my nose? Like, my ears are ringing? I don't what what happened?
Speaker 4:So, there were a couple of great, like, product usage type questions, that that came up on, on Twitter and and other places. Someone says, seems like a cool product. How does this work with software updates? Are they subscription based? Are they free forever?
Speaker 4:If the company goes bankrupt, will the software still work? Is the source code open?
Speaker 1:Yeah. I mean, a bunch of good questions in there. I mean, I would just say on upgrades. I know Emily in the chat had asked this question a little while ago. Emily, I think you got some good answers on that, but I just wanna elaborate a little bit that the asking just about upgrading the abstract.
Speaker 1:Like, how upgradable is this system designed to be? And in many ways, we viewed the ability to to upgrade as the most important feature. In fact, arguably, the only important feature. And in fact, we if you hear anyone in Oxide refer to MUPTATE, MUPTATE is the minimum upgradable product It is is MUP, and, we it is very important for us that we are able to upgrade this, and a lot of effort has been spent on exactly that. In terms of the the economic model around, I mean, you know, it's certainly it's all open for sure, and it it's all out there.
Speaker 1:I think that, you know, we would it would be certainly be our intention to empower those customers. And I think that we've always believed that you can cleanly differentiate support, from which is to say the act of calling up an engineer at Oxide and getting support on an infrastructure problem versus, availability of updates, but I don't know. Steve, you're not kicking me out of the table yet, so I thought Not yet. The item. Pretty far away.
Speaker 1:I'm I'm getting further and further away here. So it's, but, no, I think it's
Speaker 2:No. That was accurate.
Speaker 4:Are there plans to open the the web console, or will it be kept closed?
Speaker 2:We're we're not keeping it closed for any particular reason. We just we'll we're gonna look at I mean, our intent is to open up all the software that is running on the rack.
Speaker 1:Is the console currently closed?
Speaker 2:Yes. Yes. It is. Yeah.
Speaker 4:Yeah. And and it and it's closed, I think, based on, basically, historical reasons. I mean, the reason why lots of software stays closed, just because, of historical reasons and not really being sure what's in the history. But we should just burn down the history and be sure of what's in the content.
Speaker 1:Yeah. Exactly. I forget saying the thing. I should just do the thing. We're gonna, like Yeah.
Speaker 1:Yeah. We just you know, Keter's suggesting I just open up the console right now. Like, no. We, we will our our our
Speaker 2:our main focus has been just recently that we've getting getting our products to market, getting them in the hands of the first customers, getting them successful, and and it is only if we are focused elsewhere that we have not gotten to opening up all the software.
Speaker 4:And I would just say that anecdotally, you know, we do have this we do intend and and usually do open source our repos immediately, open up the repos immediately. Sometimes we keep them closed as they're kinda sketches, and it's we're not really sure what they look like yet. And amusingly, one of the precipitating factors to open sourcing these repos has been just inter repo, like, git GitHub actions. Oh, yeah. Like, oh, man.
Speaker 4:Like, have like, linking to a private repo is kind of a pain.
Speaker 1:It it is
Speaker 4:okay. Source it.
Speaker 1:Yeah. We're just open source it. I know. I well, this is one of my more viral tweets, which was totally true at the time was, like, the truth is that we we open source our software because it was easier to search the Internet than our internal Wiki, which so, yeah, there's a there are a bunch of reasons to open source software. Developer velocity is definitely among them.
Speaker 1:So, yeah, we'd look for us to get all that software out and open. And, I I guess we should do should do do a sweep to see what else. I unfortunately, I think, like, Omicron, for the reason you mentioned, I think that it's, all of that has been biased towards being open because it's so hard to build if it's if it's kinda half private.
Speaker 4:Yeah. The only some some folks in chat have noticed that a couple of networking repos are not open, and we do kinda weird things in order to build Omicron. And some of that's due to, you know, both real and perceived concerns around IP from some of our networking hardware vendors, but I think that we may be able to move past that. Some of it's just fear about what's in the GitHub history or in the Git history.
Speaker 1:Yeah. For sure. And, the, do you build your own your own repos on your own hardware? We will ex Eliana with the Ezra in the in the chat, we will real soon now. No.
Speaker 1:We, the reality is that our own hardware has been, with it has really only been in the last couple of months that we've been able to get full racks together. Just because, I mean, when you go listen to our compliance episode, if you wanna know that whole adventure, but we are, that is absolutely where we're going. So, it will be real soon now. And yeah. Exactly.
Speaker 1:And we need to get the CI times then. So, we're we're very excited to go to that.
Speaker 4:So I know there are lots and lots of, of additional questions, and we'll stick around to answer them. But I did have one question from chat that I thought would be a great one to end on, which is what part of this journey was hardest? The scope increase when you couldn't use reference designs, dealing with supply chain, raising money for this kind of start up. I thought that was a great one to to hear from from both of you, Steve and Brian, about what's been hardest about this journey.
Speaker 2:Write it down. Turn it over.
Speaker 1:Write it down. Turn it over. Yeah, baby. I I've yeah. I mean, I've I you know what?
Speaker 1:If I got the if I've got the truth lasso on me, I will describe something. I'm not even sure Steve has heard. This is I know that I that I've been so, yeah, I will describe the hardest moment. Do it. Alright.
Speaker 1:The hardest moment was when we got funding. When we got the term sheet, which is gonna this is gonna be this is gonna sound, like, absolutely unhinged. And I was not ready for this at all. I just did not. I was shocked by this.
Speaker 1:We got the term sheet, which is great. Like we've been, we've been focused exclusively on this. We've been working so hard. We've got the term sheet. We're doing it.
Speaker 1:We're done. What did you know, it's like, how would you like, that seems like great news. And for honestly, like, the I was, and I guess I was, like, not braced for this, but all of a sudden, there's this moment. I was the it was the dog that caught the car. There's like,
Speaker 2:oh Uh-oh. Shit. And but
Speaker 1:it wasn't even like it it was like worse than verbal. I mean, I I was like, I was, like, paralyzed for, like, a weekend. I was paralyzed, and I was actually were you paralyzed as well
Speaker 2:to be? I I just I mean, not not in the same level of paralysis, but you were definitely I I I recall that vividly.
Speaker 1:Totally. And in part because I knew who we just signed up for. I knew, like, all these VCs, like, oh, we don't see those technical risk here. It's like, there's nothing but technical risk. And it's like, oh, we know you can do this.
Speaker 1:Like, I don't know that I can do this. And then, like, I could just I need we need a team to go to go do this. You know, it's like, there is no way. We have so much to go do. And this is before we had our first employee our first employees, we knew we're gonna get, in you know, god bless you.
Speaker 1:But, you know, and and, like, they were, like, pretty interested in potentially coming to Oxide, but, like, they weren't at Oxide. Like, they had other jobs. And, like, I and, yeah, Robert was Robert was the first employee, and the, there was a great moment with Robert. And Robert is kinda, like, inching over the line, and Robert's like, I guess I work here now. And Steven are both like, you do.
Speaker 1:I guess you do.
Speaker 4:I guess you do. It's like, oh, a little bit of like, okay. I guess I do.
Speaker 1:Lock the door. Lock the door. The and I actually remember, like, that I have never felt as paralyzed by work to be done. And I remember I was I I I spoke with my mom, and, you know, god bless my mom, super supportive. And and my mom kinda told me what I needed to hear at that moment because I actually and I remember Steve, I I did this, like, you and I have I have confided basically everything in one another.
Speaker 1:But this was a moment where I'm like, I really you you don't want your, like you know, I I just felt like it would be contagious if I went to you and I was like, hey. Look. So, I'm terrified. You'd be like, you what?
Speaker 3:You I I I was
Speaker 1:I was not expecting. Right? I I assume we weren't gonna get the money, and now I don't know how to build it. Steve, I don't know how to build it. You'd be like Do you know how to build a computer?
Speaker 1:Yeah. Do you have a computer? Because I don't know how to build a computer. You'd be like, what did I just do? Are you kidding me?
Speaker 1:And so I was a little bit like, I cannot I there's a degree to which I and I clearly I didn't totally keep the straight face because it was obviously, it's
Speaker 2:not Not to that degree though.
Speaker 1:But, but it got a little bit dark. And, and my mom was like, you know, you, you did it before. I'm like, wise. Definitely wise.
Speaker 2:And she said, well, think about what you
Speaker 1:did at Fishworks. But you think about what we did at Fishworks when we were a startup inside of Sun Microsystems where it's like, no. Not the same. And she's like, well, just think about like, you can do it. I'm like, mom I I remember thinking, like, mom does not know what she's talking about because she has got no idea of how much needs to be done here.
Speaker 1:And, you know, Adam, you had said this when when I I was trying to when trying to convince you to come to Oxide, and, you know, you the way you you, characterized it was, this is Fish Works without a net, and that is what it was. It felt like Fish Works without a net, and I was really focusing on the absence of the net at that moment. I'm like You were looking down. Yeah. I'm looking down, and I think that I what I had not done is the thing that I had learned early on to do a joint, and I would advise anyone to do going to a start up.
Speaker 1:When you go to a startup, success is not assured, and you'd, like, you kinda need to mentally come to come to grips with the death of the company. And the which is important, like, if the comp because if the company capsizes, it's gonna be okay. And, like, I found it coming to joint where we learned that very early on. I kinda came in and be like, this is a stable company. And then, like, on day 2, you're like, this the wings are on fire, and the the the ground is coming up really quickly.
Speaker 1:No pilot. There's no pilots. Exactly. The you know, the pilots are are in a slap light with one another. And, you know, it's like one of the pilots is drunk.
Speaker 1:Oh my god. They're actually about both drunk. The, the and so, like, being able to come to grips with the fact that, like, okay. Actually, joint, it it would be okay. And that actually allowed me to focus on what I need to focus on, which is go build the team and and go do what we need to go do.
Speaker 1:And as a result, like, we didn't, like, hit the ground. Like, we actually did and success welcome. And I kinda feel like I ultimately it's like, that wasn't the case here. It wasn't like, oh my god, oxide's gonna die. It's more like, oh my god, there's so much to go do, and I don't where do I start?
Speaker 1:You know, where do I start? And I think that, you know, for those of you have, have come to oxide, I think that like, there's been a variant on this problem that lots of people at Oxide have had, that peo everyone kind of romanticizes the greenfield, the clean sheet of paper. And it's, like, for good reason. Like, it's it there's elements of which it's, like, pretty great. It's also terrifying as hell.
Speaker 1:And when you are looking at a blank sheet of paper, you're like, where do I start actually? And, you know, ultimately, obviously, got over it and and got my legs underneath me. And in part because we you know, thank you, Robert, as employee number 1. But, also, thank you to the all the you know, as you I think Aaron Levy had this line about, you know, before we had 10 employees, like, every employee was this it was so hard to get every employee in there. And then when he gets that the you get that 10th employee, and all of a sudden, it feels like, oh, this is like this is not a suicide mission.
Speaker 1:This is, like, actually, like okay. This is, like, actually, like, a real company, and it feel it has this feeling of stability.
Speaker 4:And then
Speaker 2:it's 20. You're like,
Speaker 1:you're 20. You're like, you're huge. I bet yeah. So the I think it becomes a lot easier to get new employees in the door. But then I but I I think that that that challenge of the the daunting challenge of the clean sheet of paper is not one that really goes away.
Speaker 1:And I know that for a bunch of folks that came to Oxide and would have this moment of, like, I'm at my dream job. Like, why am I feeling, like, so overwhelmed by this freedom that I have always wanted? And I'm like, I have been right there with you. I know exactly what you're going through, and it's gonna pass. And it's like you because I think that that's the feeling you've got is, like, it's not gonna pass.
Speaker 1:I've forgotten how to fly. I don't know how to fly an airplane. You know? We're taxing down the runway. I don't know what the dials do.
Speaker 1:I just like, we're all gonna die. It's like, oh, but that the that's not the gas light. It's the intercom light. That's like, okay. They and so I think that that for me so, Adam, long answer, but that's that that was the lowest moment.
Speaker 1:I think every other there have been plenty of bullets over the ear, for sure, and plenty of crises, but we have endured so many of them together that it's just like, I I I'm just ready for anything at this point. I'm just like not this is why part like, my anxiety level has actually gone down over time. Even as, like, sometimes, like, the stakes have gone way way up because of the team working together. So you go, Steve. That's my long answer.
Speaker 1:Good.
Speaker 2:It's good. I'll I'll I'll try and make this one a little shorter. But I think the I mean, raising raising for a company that also builds hardware in the pose in the in the cloud SaaS era was was certainly difficult. That that that was a challenge. I think the I'll give it I'll give two answers.
Speaker 2:One is sort of the overall what was what what has been one of the more difficult things in the past three and a half years. And I think my answer to that would be getting buy in from the companies that we would need to rely on to build the system we were gonna go build. Totally. So, like, AMD comes you know, AMD is an obvious one. But, you know, when we start out, we are that exact startup, that half paralyzed startup that you just described.
Speaker 2:And we are trying to not only get the attention of, but, like, punch way above our weight, which was, you know, 2 ounces at the time. Like, we we to be able to get access to the resources that we would need, to be able to build this system. And, you know, one of the things we talk about is that the riskiest one of the riskiest parts of what we were gonna go set off and do was to do our own firmwares, to rip the bios out of the system and go write our own kind of bespoke, systems layer of software, and we needed documentation. We needed Yeah. You know, we needed a lot from, you know, folks in the industry that don't always document things the best way.
Speaker 2:And, so being able to to get the level of resources we needed from a number of companies that we couldn't build the system without was, was stressful. But singular moment, just back to you using a singular moment Yeah. That was the hardest, for me was was relatively recent. It was actually this year. And it was, around the banking crisis.
Speaker 1:It was. The SUV failure.
Speaker 2:And it was when the bank that we banked with failed. And it it wasn't it it was not that I was afraid or or over I mean, this is not to say I was not stressed out over the over the the mid term the medium term impact to oxide. I had absolute conviction in the longer term oxide because we had our, you know, terrific investors that were coming in saying, like, we're we're going to make sure that you all can endure and that we're gonna make it through this. I think it was the hardest thing was all of her all the folks at Oxide that had all of these very important questions that they wanted to get answered around what's happening.
Speaker 1:Yeah.
Speaker 2:What's gonna happen with our money? What does this mean for payroll? Does it mean for, you know, the next 3 days, 3 weeks, 3 months? And them getting all those same questions from their family members and their friends.
Speaker 1:So was my very low anxiety over this, was that just was that, exacerbating? Or was that was actually helpful or not helpful?
Speaker 2:I mean, I don't know.
Speaker 1:So I mean, I have said at the time, and I still believe that, like, the Silicon Valley Bank failure is just like just doesn't feel like I was saying top 15, and I'm saying top 50 existential threats to
Speaker 2:oxides. It wasn't the it wasn't the failure itself. It was just, again, not being able to help folks at Oxide be able to answer these questions. And there there are levels of stress that was hard to reduce. And that's right.
Speaker 2:Situation over the course of 4 days, that was that was that was toughest.
Speaker 1:It was so it was actually the team itself
Speaker 4:Yeah.
Speaker 1:More than the like, you have got your own confidence that, like, yeah. We're gonna figure we are gonna navigate.
Speaker 2:Knowing that everyone on the team was
Speaker 1:Was dealing with their same anxiety. That's interesting.
Speaker 2:Sky high level.
Speaker 1:Yeah. Pathetic. Yeah.
Speaker 2:Not no way to really diffuse that. Over that period of time in which everything hung in the balance Thursday, Friday, Saturday, when, like, is SVB gonna survive? Are they gonna get bailed out?
Speaker 1:Right.
Speaker 2:And the outcome of that was gonna have sort of a a pretty different set of paths for us over the next 6 to 12 months. Yeah. But, again, with conviction, we were gonna make it through, and we we were not gonna, you know, be overly affected. That that inability to kinda help our team Well, I deal with that anxiety.
Speaker 1:So and part of this, my anxiety was so low. I know I've said this before, but it's, like, Janet Yellen does not care about any of our other problems. Janet Yellen will gladly, like, have like, a Chelsea or Nick that can't actually, like, come out of reset, Janet Yellen does not care. We will we will die in the wilderness alone. And I'm like, hey, SVB.
Speaker 1:Like, Janet Yellen dialed in on this. It's great. Like, I you know, we got so many people care about our problems. This is really terrific and refreshing. I'm so used to being, like, totally in the wilderness.
Speaker 1:But I yeah. The that point about, like, actually, the problem is not that. Like, you got confidence. Like, we're gonna get it feel inside there. The problem is you've got a lot of employees who are like, hey.
Speaker 1:Like, you know, my mom just asked me who my bank we banked with, and my dad just asked me who we banked with. And I'm like, do we we don't bank with SVP, do we? It's like, no. We no. We a 100% do.
Speaker 1:But, like, but we've got some stuff elsewhere. It's like, we sure don't. It's like, okay. Well, what's the good news now? What's the insight you or we're we're paying attention to it.
Speaker 1:And it's also we do we've got actually on the one that good news Janet Young cares, bad news, we ourselves had very little agency in that moment. That's right. Yeah.
Speaker 2:So we're on the other side of that. We are in we're, you know, financially in in great, great straits for years to come, and all's well on that front. But it was that was particularly Low moment. Low moment, but I think, you know, on the partner front
Speaker 1:Potential investors that treated us shabbily, you have all gotten off easy. That's right. Well,
Speaker 4:it it was a great answer and in particular coming full circle from when deflecting, kind of praise to the team to, probably, you talk about some of the anxiety of building the team and Steve taking care of the team. I just I love that, and I would say as part of the team that we really feel that every day.
Speaker 1:Well, I it is ultimately it it it, emphatically, it is the team that has done this. And that's part of the reason I think that, you know, I I just said the jump. Adam and Steven are kind of squirming them comfortably in our chairs when you're talking about us having done this because we really view it as it is the broader us have done it. And, the I think the the only thing that we can really claim credit for is for the initial raise and bringing in the initial folks, and the the rest has has been, a a terrific glorious flywheel, and it is very much included those who are fans of oxide. You know, I think early on, one of the things we did is we asked people, hey.
Speaker 1:Like, if you want a sticker, fill out this Google form. If you want a sticker, we'll mail you a sticker. And, Adam, do you remember that we had, like and this is, like, before the pandemic, and all of a sudden, we had, like, like, like we got stuff like 500 envelopes full of stickers. And I it'd be, we ended up, you know, my kids, my mom, my mom was stuffed in envelopes, and and really the time thinking, like, man, this is how great is it to have a, a just a broader set of folks who are rooting for us. So, it has really meant a lot for us, for us collectively, us all.
Speaker 1:And so thank you for all of you who've been who've been supportive of us. And
Speaker 2:didn't think we were nuts.
Speaker 1:And didn't think we're nuts. Absolutely.
Speaker 4:And maybe just the supportive one. We all thought we were nuts.
Speaker 1:It's true. I know.
Speaker 4:So in in, next week from, you want Brian, you wanna tee up,
Speaker 1:the next week is gonna be next week's gonna be amazing. So we're gonna have the operations team on here. In particular, we are and we're gonna have a bunch of folks up from operations. In particular, Eric Anderson's gonna be here. Eric has been more or less on-site on our facility, in Rochester.
Speaker 1:And Eric has got some just terrific insight, into, all of the gritty details to manufacture this thing and all the the terrific collaborative work with our our contract manufacturing, Benchmark Electronics has been terrific. And really, I'm really looking forward to having Eric and team. CJ is gonna walk us through. I'm hoping to see we get CJ to walk or CJ or Kirsten to walk us through the crate. Kate's gonna be here.
Speaker 1:It's gonna be really terrific. So I'm really looking forward to having the, the ops team in next week.
Speaker 4:Thanks, guys. Thanks everyone in chat. This is fantastic. I have so much fun.
Speaker 1:Awesome. Thank you, Adam, and thank you especially to all of your alts.