Today marks a big announcement. The Aurora supercomputer at Argonne has its compute blades installed. We have been tracking the Aurora supercomputer for years, and we expect it to dwarf the current HPE and AMD Powered Frontier first Top500 Exascale machine once it is launched.
Argonne Aurora Supercomputer Intel Max Blade Installation is Complete
Previously, Intel said it would be installing its CPU and GPU blades with Intel “Sapphire Rapids” CPUs, but without Intel Xeon Max HBM CPUs. The Xeon Max CPUs would be retrofitted at a later date. We sent a question to Intel yesterday to confirm that this milestone includes the Xeon Max CPUs with HBM onboard, and Intel confirmed this was the case.
Today’s announcement is that all 10,624 compute blades each with two Intel Xeon Max CPUs and six Intel GPU Max GPUs (codenamed Ponte Vecchio) are now installed. That means there are a total of 21,248 Xeon Max CPUs and 63,744 Intel GPU Max series GPUs installed.
The Ponte Vecchio chip is perhaps the most advanced chip ever made integrating dozens of chiplet tiles from different fabs. That is why we called Intel Ponte Vecchio a Spaceship of a GPU. It is an accomplishment that Intel made over 60,000 of these GPUs that actually work.
The Aurora supercomputer is using HPE Cray EX liquid cooling and a massive Slingshot interconnect system that is also powering the US’s first three Exascale supercomputers. Last year we took A Look at a 2023 Top500 Aurora Supercomputer Node with Intel Ponte Vecchio in more detail. We also looked at the CoolIT lab working on the liquid cooling for HPE Cray EX systems in How Liquid Cooling is Prototyped and Tested in the CoolIT Liquid Lab Tour.
Future supercomputer systems we expect to all be liquid cooled.
Final Words
ISC 2023 this year was a bit of a bummer. We did not get a new #1 on the Top500 list. Last week we covered the AMD Instinct MI300 which will find its way into 2+ Exaflop systems. This Aurora system is going to eclipse Frontier when it is online. Given today’s progress announcement, we hope that means we get a new Top500 #1 supercomputer at SC23 in November.
This is a big milestone, now the wait for the system to run linpack and show its performance. In the meantime, we will be showing you some cool features with Xeon Max on STH.
Obligatory can it run Crysis?
If the Optane DAOS on Aurora turns out to be a key factor of the “perfect environment”, as Rick Stevens has called it, for the science AI model training, I think Intel is going to have to re-evaluate pulling the plug on that program.