Nvidia has announced that its DGX H100 computers will use Intel Xeon Sapphire Rapids processors. The green company is switching from AMD’s EPYC processors to Intel’s Xeon processors after several years
DGX H100: Nvidia chooses Sapphire Rapids over AMD EPYCs
Jensen Huang claims that the main reason for the move to Intel’s Sapphire Rapids is their exceptional single-thread performance, which seems to far outperform current EPYCs. Intel appears to be shipping its new Xeon CPUs to customers, including Nvidia. AMD has yet to release its EPYC processors with Zen 4 cores that will compete with Sapphire Rapids. However, Nvidia can’t wait and has decided to go with Intel’s proposal.
Sapphire Rapids is Intel’s next-generation server architecture based on the current Alder Lake Core desktop and notebook processors from the 12th generation. In other words, Sapphire Rapids is the server version of Alder Lake.
So the DGX H100 computers would benefit from Alder Lake’s cores and all the innovations like DDR5, HBM2E, and PCIe 5.0 support. The total core count per unit offered by Sapphire Rapids is 56 cores with support for up to 80 PCIe 5.0 lanes.
The green team would prefer the single-threaded performance of these processors over multi-threaded performance, where AMD has an advantage by offering more cores per EPYC unit, such as 64 cores versus the Sapphire Rapids’ 56 cores.
The DGX H100 computers are designed specifically for artificial intelligence applications, and each features 8 of Nvidia’s new Hopper H100 GPUs with a performance of 32 PetaFlops.