The Oxide and Friends revive an old Solaris Kernel Group tradition of making predictions, Adam and Bryan are joined by Redmonk's most famous Red Sox fan, Stephen O'Grady, to make 1, 3, and 6 year predictions. Kelsey Hightower stops to dispense some present and future wisdom.
Oxide and Friends Twitter Space: January 3rd, 2022
47:15 | Discord are going to annoy their userbase. | We’ll finally get a RISC V server in a datacenter, in some shape or form. | Email goes the way of the landline.
1:10:05 | The framework laptop company will be unsuccessful. Existing laptops are not substantially different; with some retooling. | One major FPGA vendor will have a completely open toolchain for high end FPGAs. | At least 1 RISC-V supercomputer in the Top 500.
1:16:45 | At least one of the hyperscalers will become startlingly good at partnering. | Stablecoins will become regulated. | The biggest datacenter server provider (outside the hyperscalers) will be a company that hasn’t yet shipped its first server.
1:18:57 | Multiple companies will have demonstrated a AGI (one shot machine learning system). It’s not gonna be useful for anything, but I think the problem is less hard than many critics think it is and several companies/organizations are actually going to be showing the first versions of these systems. | Drones autonomously flying around private properties will be a common thing. Factory managers, powerlines inspectors, large building sites etc. will have commonly available and affordable options to inspect or patrol their properties. | Web3 will actually happen, but not in the way it’s currently being talked about. In 6 years time bots will have improved to the point that they can not be warded off the major platforms (or any platforms) and will make the web absolutely unusable due to them disrupting all established crowd funded moderation systems. A new paradigm will have to emerge that fundamentally changes how we use the web (thus web3), so that we can still derive value from it.
| Ben Stoltz
1:24:40 | Smart glasses become a viable alternative for computer monitors youtube. People who used to look away from their phones to have their own thoughts, and are now using smart glasses in real life situations, are subjected to an ads vs. attention “Tragedy of the commons”. As costs per unit decrease leading to ubiquity, this forces a modern-day “Highway Beautification Act” to legislate Ad Blocking. | A significant percentage of commercial office space will be converted to housing. | The best AIs have emotional problems. We don’t really know how they work. AI specialists are more therapists than programmers.
1:29:30 | This year will be more of the same, competition to define the new normal as the pandemic winds down. | Pandemic-era solutions will backfire; crypto-currencies will give governments an excuse to track all actual spending. “We will give you the transparency, but not the kind you wanted.” | Technology will be recognized as sovereignty like money and land used to be. Governments will be wary of using technology from weak allies or competitors. Local hardware manufacturing, growth of local university training, etc. Possibly manifesting as national protectionism, or a reprise of the space-race. Open source will be the default model.
1:53:45 | a major OS from China emerges | high performance computing from Europe | ARM no longer as relevant
1:58:00 | web3 is done; we’re not talking about it, it’s not a thing, we don’t use the term and we only vaguely recall what it was supposed to mean. | Productivity per watt becomes a highly important metric in computing. Tools tell us about our power use. We spin workloads up and down depending on power cost and availability. | AWS offers RISC-V instance types.
1:07:14 | Single-node computing: people will realize that that distributed computing has a lot of overhead and that one server can do a lot of work. This will lead people to people doing business analytics jobs by pulling all their data to a single a computer and doing the calculation, getting the result 100x faster than splitting data over many computers. | Microservices inlining: taking a lot of microservices and statically linking them together. This will enable calling functions without network overhead, making things run 100x faster. | We will start do scaling properly. Instead of thinking “how can I make this big data and scale up to infinity”, we will try to get the most out of single node. Only once a single node has been pushed to its limit will we scale up to first a rack, then a datacenter, and then the world.
2:01:10 | Major workplace changes due to the pandemic will amplify and accentuate the wealth gap and disparity. Only some industries are privileged enough to be able to work from home. This will create social problems. | Regulation of social media in the aftermath of widespread political unrest, particularly after the US 2024 political season. | The effects of climate change will be sufficiently apparent that people will get serious about retooling around compute and power efficiency.
56:06 | No one year prediction. | CCPA copycat laws in other states, perhaps US federal legislation, plus changing global regulatory environment lead to GDPR-like protections to no longer be geo-fenced by bigger players. This’ll also have impacts on SaaS adoption - spreading data around makes right to amendment and right to deletion harder. | RISC-V chip in mainstream phone (likely Samsung). Previously moving target, but longer upgrade times and slower pace of improvements will cause Samsung to chase RISC-V for high volume phones due to better unit economics. Will have prior experience in RISC-V fab for other applications.
59:50 | Not much will happen in with tech industry anti-trust. Any bills in progress will be watered down. | Antitrust enforcement will have a chilling effect on acquisitions. | Section 230 of the Communications Decency Act lives. The ARM/NVidia merger will not go through, but it’ll will take 5 years to find out.
Prediction about optimizing compilers | We’ve reached the point of diminishing returns in existing LLVM optimization passes (<3% headroom, which is hard to get). | Compilers not only help optimize code, which is what they traditionally do. Compiler developers shift their focus on how they can enable application developers to write more efficient SW, i.e. help them indirectly. For example, a recommendation system based on the database of “golden” code written by performance gurus. | Machines are capable of programming themselves. Humans just feed a (formal) description of an algorithm to a machine and it comes up with an optimal code tailored to this particular HW. Compilers play an essential role here. (This is more of a 10-year prediction)
| You? | | |
What is Oxide and Friends?
Oxide hosts a weekly Twitter Space 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 Mondays at 5pm PT for an hour or so to catch us live.