After almost a year on the market, Cray has adopted AMD EPYC for its high-performance computing platforms. The Cray CS500 supports AMD EPYC CPU blades for compute-intensive tasks. This is a big deal for both Cray and AMD as each has something new to offer. For Cray, this is an alternative-to-Intel x86 architecture. We see Cray adding more types of CPUs to the mix soon but diversification is happening. For AMD this is a big deal as it has what is perhaps the most storied supercomputer brands asserting that EPYC is ready for HPC workloads.
Cray CS500 with AMD EPYC
The Cray CS500 cluster solution is extremely flexible. It offers the ability to scale to over 11,000 nodes in a single cluster. Cray offers not just the hardware and interconnect stack, but offers cluster management and application platforms as well. Here is the Cray cluster software stack.
In terms of compute options, the CS500 series offers Intel Xeon Scalable, AMD EPYC and Intel Xeon Phi for x86 compute, NVIDIA Tesla GPUs for GPU compute and also Nallatech FPGAs for acceleration. Fabric options range from Omni-Path, Infiniband, and Ethernet and can be configured with options for 3D torus or single/ dual-rail fat tree. For storage, Cray uses fiber channel attached arrays and can support SSDs and disks in the arrays and also as local storage. File system options include Cray ClusterStor, NFS, Local FS (Ext3, Ext4 XFS), Lustre, GPFS, and Panasas PanFS as global file systems.
This is a big deal for AMD EPYC. Being part of a Cray cluster solution means two things. First, to the industry, AMD is able to show another well-known traditional vendor supporting its product. Second, companies like Cray often develop these solutions based on customer demand. The fact we are seeing a publicly advertised solution means that there is a good chance one or more customers are actively seeing AMD EPYC in the HPC market today.
You can learn more about the Cray CS500 series here.