Show Notes
In this week’s episode of Hewlett Packard Labs Podcast “From Research to Reality”, Dejan Milojicic hosts five colleagues across HPC/AI business and Hewlett Packard Labs: Nic Dube, Paolo Faraboschi, Arti Garg, Ken Leach, and Rangan Sukumar. Together they present a point of view on heterogeneous computing and they issue a call for action. Outside of the embedded space, the computing industry is unprepared to deal with a flood of new, specialized and unconventional computing elements. They argue that the community needs to embrace a major shift to enable heterogeneity in an equitable, secure, and sustainable manner. This shift has to cover every aspect of how we design systems, including platforms, interconnects, software, applications, and IO subsystems. With the increasing innovation in the accelerator space and several new accelerators entering the market, there currently is a matching and unjustified diversity in the systems built around them, software, and management stacks. While in some cases system-level diversity is justified by a different computing paradigm, the community needs to avoid “gratuitous differentiation”, which does not add value to the diverse computing elements, complicates the integration process and application portability, and causes a multiplication of efforts that shifts engineering resource away from core technology. As a result, end users are facing increasing difficulty in reusing a single application code base across different accelerators. Similarly, system integrators are facing increasing difficulty in incorporating accelerators in their products, and software developers in developing code that is portable across multiple heterogeneous computing elements. To overcome these problems and to encourage focused innovation in the accelerators space requires understanding the system-level differentiation points and providing a set of recommendations to all the participants in the ecosystem, including accelerator technology providers, system integrators, software developers, the open source community, as well as the end users in both the scientific community and the cloud service providers. The goals of these recommendations would be to identify the areas to open vs those where innovation is welcome. This document motivates this point of view on this topic, but also represents a “call for action” for all the major players to work together and improve the landscape of heterogeneous computing. This is an important next step that will benefit the entire industry, and collectively improve the ability to achieve faster results and better reuse across accelerated systems.
What is From Research to Reality: The Hewlett Packard Labs Podcast?
This is a podcast about the many phases of technology innovation: ideas, research, development, management, marketing, and many others. We’re talking to a wide range of people with diverse and rich backgrounds including all species of engineers, scientists, mathematicians, business people and technical managers. We hope you will enjoy these podcasts as much as we did making them.