AMD Announces Milan-X 3D V-Cache CPUs, Azure Prepares For Great Upgrade

AMD this morning hosted their Accelerated Data Center Premiere virtual event where they introduced Milan-X as well as the Instinct MI200. With Milan-X there is now 3D V-Cache introduced for the current generation and still-very-impressive EPYC 7003 "Milan" processors. I can personally attest to Milan-X being very exciting for HPC workloads and beyond with impressive gains to performance.

Before getting to the CPU fun, AMD did announce today the Instinct MI200 series. MI200 is what was developed over the past year with the Linux driver support as Aldebaran.

The AMD Instinct MI200 series is their first CDNA2 product and offers up to 220 compute units, up to 128GB of HBM2E memory, up to 880 2nd Gen Matrix Cores, and is comprised of up to 58 billion transistors and manufactured on a 6nm process.

With AMD's numbers, they are showing the MI200 slaughtering the NVIDIA A100 accelerator.

AMD did also tease ROCm 5.0 is on the way... Besides adding MI200/Aldebaran support there, their open-source compute stack will hopefully see official RDNA/RDNA2 GPU support with ROCm 5.0.

For those eager to see AMD processors with V-Cache, while Ryzen CPUs with V-Cache are reported for Q1, today AMD is rolling out 3rd Gen AMD EPYC CPUs with this 3D chiplet technology.

AMD EPYC with 3D V-Cache can offer triple the amount of L3 cache over existing EPYC 7003 series processors with 804MB of cache per socket! Yes, up to 1.6GB of cache per EPYC 2P server! These 3D V-Cache enabled processors still support up to 64 CPU cores as with EPYC 7003 and remain SP3 compatible with all existing EPYC 7003 series motherboards. Only a firmware/BIOS update is required for supporting these new processors on existing platforms.

With this huge amount of cache per processor, AMD is reporting a "50% average uplift" across targeted workloads. Among those targeted workloads are finite element analysis, computational fluid dynamics (CFD), electronic design automation (EDA), structural analysis, and other HPC workloads. Basically Milan-X will perform the best for very large workloads where not all of the data could previously fit well into the L3 cache.

While being announced today, the AMD 3D V-CACHE EPYC processors are not officially launching until Q1 2022. Up next is how Microsoft is making Milan-X even more interesting with their Azure HBv3 upgrade.