Supercomputers aren’t exactly high-volume machines since they tend to be one-off efforts custom-designed for a particular customer. But there is a lot of effort put into assuring a certain country or vendor become the fastest and the US is a race with China, and HPE and other vendors are in a constant race to see who can build the most powerful Supercomputer. The winner not only gets bragging rights, but they’ll also get additional business because, at this scale, time is very costly. HPE and Cray (Cray was, at one time, famous for being the leader in supercomputers), merged some time ago, and this latest effort to create a Supercomputer will be a showcase of what the combined company can do. Called the El Capitan after the mountain in Yosemite National Park, one of the most famous mountains in the US, and targeted at problems for the US Department of Energy, NASA, and the Lawrence Livermore National Laboratory promises to be a huge game-changer once put into service in 2023. Let’s talk about that this week.
Incredible Power At Scale
This computer is a showcase for both HPE/Cray and AMD because it uses a blend of AMD’s CPU and GPU technology, and wrapped with AMD’s 3rd Generation Infinity architecture. This system will be a showcase for AMD’s coming “Genoa” architecture, which will be a game-changer. Genoa is a Zen 4 processor with next-generation memory and I/O Sub-Systems designed for Artificial Intelligence and High-Performance Computing workloads.
The Supercomputer will focus on jobs like modeling and simulation, Artificial Intelligence, and Big Data Analytics at Exascale levels. The performance will be two-exaflops or two quintillion mathematical calculations per second. If you don’t know what that is, we have thousand, million, billion, trillion, quadrillion, and then quintillion, which has 18 zeros, which is why I’m not typing that out. That’s a lot of fricken zeros. The result should not only be 10x faster than the fastest Supercomputer today; it will be faster than the top 200 supercomputers combined. For instance, if everyone on earth did one calculation per second, it would take them eight years to do what this computer can do in one second.
This kind of power should make it much easier to make high-quality weather predictions, create far more accurate large scale 3D simulations, and it will clearly break the existing Exaflop barrier and vastly improve the frequency and quality of results that NASA currently gets from existing supercomputers.
Importance For AMD And HPE
A supercomputer at this scale showcases AMD’s ability to create very high-performance server solutions that not only outperform competitors but do so cost-effectively. These systems are bid on both price and performance with multiple validation steps during the development and deployment of the end offering. Vendors are given little leeway in terms of missing performance goals, so the system will have to perform as advertised.
Other Server OEMs see this and take it as a validation step for the underlying products and architecture, in this case, AMD’s Shasta architecture. This Supercomputer is more than bragging rights; it is a real showcase of what AMD and HPE can do if they put their minds to it and the computer, once it launches and is validated, should be a huge asset for both companies as a showcase.
HPE has been having a tough time of late, and it is critical the firm showcase that its problems aren’t adversely impacting its execution. So it will be critical that HPE/Cray meet their design goals and timeline, which, I expect, will do. The result should showcase that you can trust HPE and AMD on the most critical projects opening up more of the Exascale market to both firms.
Wrapping Up: Setting The Supercomputer Record And Redefining The Future
This HPE and AMD effort will provide a level of performance we haven’t been able to approach before but do recognize this will be a race, and until this system ships, someone else could step in and steal the crown. With a 10x advantage that does seem unlikely, but until they reach the finish line, it remains anyone’s race.
We win regardless because this kind of performance will be critical to anticipating the increasing number of extreme weather events, calculating how we go back into space, and modeling the world of the future. It might also help us find the next anti-virus and develop the first general-purpose AI. The future is coming, and the El Capitan Supercomputer will have a lot to say about what it looks like, and it will certainly help both HPE’s and AMD’s images.