This is a big win! Previously AMD had joined with Cray to build a pre-exascale supercomputer named Perlmutter with funding from DOE. Perlmutter will use AMD’s Epyc 3 (Milan) CPU and Nvidia’s next gen GPU in 2020. It was to prepare for an eventual exascale supercomputer.
Fast forward to today, amazingly, DOE is now asking AMD and Cray to build a 1.5 exaFLOPS supercomputer, named Frontier, using AMD’s next gen customized Epyc AND AMD’s next gen customized Radeon Instinct GPU to power the fastest supercomputer in the world in 2021! This means that Perlmutter is coming along nicely, and also that AMD has beaten Nvidia in the GPU front as well.
Concurrently, another exascale supercomputer will use Intel CPU and GPU (still in development) which will (only) be capable of 1 exaFLOPS.
Nvidia is out of the loop! AMD and Intel are favored because they both can provide both CPU and GPU. All the supercomputers in the Top500 list have pretty much deployed all Nvidia’s GPUs. See below for the latest list for the top ten.
https://www.top500.org/lists/2018/11/
This is a great vote of confidence from DOE. The halo effect is priceless. For reference, the current fastest supercomputer in the world checks in at a mere 0.2 exaFLOPS.
[…] Frontier (1.5 exaflops) was award to HPE/Cray, using AMD’s custom Milan CPU plus Radeon Instinct, for Oak Ridge National Lab, detailed here. I had also written about it, here. […]
LikeLike