Four AMD EPYC powered supercomputers are listed among the 50 highest-performance systems in the world and was announced by the company itself, today. This includes the seventh spot on today’s TOP500. In total, there are 10 AMD EPYC-powered supercomputers in the list. The momentum for AMD EPYC processors in advanced science and health research keeps growing with new installations at Indiana University, Purdue University and CERN, as well as high-performance computing (HPC) cloud instances from services such as Amazon, Google and Oracle Cloud. According to Forrest Norrod, the leading HPC institutions are fully utilizing the power of 2nd Gen AMD EPYC processors to cope up with the extensive task done by the scientists.
Below are the list of supercomputers powered by AMD EPYC that in the list of TOP500:
- Selene (No.7) an AMD EPYC 7742-based system in a DGX A100 SuperPOD platform from Nvidia,
- Belenos (No.30), one of the two BullSequana XH2000 supercomputers at Météo-France, the French national meteorological service powered by 2nd Gen AMD EPYC processors
- Joliot-Curie (No.34), moved up the list based on a new submission for the BullSequana XH2000 system using 2nd Gen AMD EPYC™ processors at GENCI, the French national high-performance computing organization
- Mahti (No. 48), a 2nd Gen AMD EPYC processor powered BullSequana XH2000 supercomputer at Finland’s Center for Science Information Technology
Powering the upcoming world’s fastest exascale supercomputers, Frontier and El Capitan, the advanced scientific and health research requires extensive workload support especially in the cloud, higher core count and extensive memory bandwidth of AMD EPYC processors are further serving the requirements from the HPC providers by improving performance, scalability, efficiency and total cost of ownership of the supercomputers.
In addition, two universities announced new research supercomputing system powered by AMD EPYC processors in Dell’s EMC PowerEdge servers. According to the source, Indiana University will further deploy Jetstream 2, “an eight-petaflop distributed cloud computing system” powered by upcoming 3rd Gen AMD EPYC processors. According to them, this system will be used by the researchers in a variety of fields such as AI, social sciences and Covid-19 research. This made possible by viewing the performance of the AMD EPYC that already being used in their Big Red 200 at the facility. On the other hand, Purdue University announced that, they will deploy Anvil, a supercomputer powered by the next generation of AMD EPYC processors that cope up with advanced computing capabilities to support a wider range of computational and data-intensive research.
As the demand increases from the leading HPC companies, cloud providers continue to adopt 2nd Gen AMD EPYC processors to provide performance and flexible solutions to meet the agile requirement that changes from time to time. AMD and Microsoft Azure continued building their own cloud-partnership via the recently announced HBv2-Series VMs for high-performance computing workloads. According to AMD further, the 2nd Gen EPYC processors provide Microsoft Azure customers with an impressive core scaling, utilizing maximum memory bandwidth and are the first x86-architecture server processors that supports PCIe 4.0, that grants high performance computing experiences in the industry. This advanced processor will be used to support real-world workloads such as weather simulations, reservoir modelling, rendering, seismic processing and many more.
AMD community support spontaneously grows for AMD Radeon Open eCosystem (ROCm), an AMD’s open source foundation for heterogenous compute, this includes six major development milestones in the latest update:
- The HIP-Clang compiler is now up-streamed and reviewed by the LLVM™ community, providing a better open source experience for the developer,
- A new rocprofiler start/stop API to enable/disable GPU kernel HSA dispatch callbacks, increasing developer productivity and shortening the profiling run times,
- AMD RCCL compatibility with NVIDIA Communications Collective Library (NCCL) v2.6.4m,
- MIOpen provides an optional pre-compiled kernel package to reduce startup latency,
- A new CPU Affinity API is introduced for aiding applications to select the appropriate memory node for a given accelerator (GPU) and a given CPU,
- The new Radeon Performance Primitives library is a comprehensive high-performance computer vision library for AMD (CPU and GPU) with the HIP and OpenCL backend.