AMD showcases Epyc HPC, cloud momentum at SC19
After a year of hitting some significant HPC milestones, AMD this week is showcasing its latest momentum in the high-performance computing market. At the SC19 conference, the chipmaker is announcing new cloud instances from AWS and Microsoft Azure based on the Epyc processor, as well as other Epyc wins and the launch of ROCm software version 3.0. AMD is also announcing the first Epyc processor-based system to join the TOP500 list.
The SC19 conference serves as a "stepping stone where we're staring to bring technologies that will power the exascale era to the market," Scott Aylor, general manager of AMD's Datacenter Solutions Group, said to ZDNet.
The "North Star" directing AMD's technology roadmap, according to Aylor, was the federal government's decision to invest in AMD technology to build the Frontier exascale supercomputer. The $600 million exascale system -- built on AMD's Epyc CPUs and Radeon Instinct GPU processors, as well as Cray's Shasta architecture and Slingshot network -- is set to become the world's fastest supercomputer when it arrives at the lab in 2021.
Then this past August, AMD rolled out the second generation Epyc, the world's first 7-nanometer server processor. The processor's performance gains that have set AMD up nicely for more HPC momentum.
Illustrating its Epyc momentum, AMD said Microsoft Azure is building its highest-performance HPC instances around the second-gen Epyc processor. Azure's new HBv2 virtual machines for high-performance computing, based on the AMD EPYC 7742 processor, are now in preview.
- AMD meets Q3 earnings expectations
- AMD's 2nd-gen EPYC continues epic market momentum
- AMD courts Wall Street, financial services firms HPC workloads with Epyc server processor
- AMD's 2nd Gen Epyc processor aims to set a new data center standard
- AMD debuts the Epyc Rome server processor
Additionally, Amazon Web Services is expanding its use of second-gen Epyc processors, announcing the upcoming availability of new C5a compute-intensive instances. That includes C5a, C5ad, as well as bare-metal EC2 instances. "They're putting their highest-performance C series instances on AMD technology," Aylor said.
AMD's OEM and ODM partners are also expanding the second-gen Epyc's reach:
- In addition to HPE's ProLiant DL325 Gen10 and DL385 Gen10 servers, the new Gen10 Plus models now feature the processor.
- The new Penguin Altus XE4218GT uses the second-gen Epyc and AMD's Radeon Instinct MI50 accelerator to power machine learning, big data analytics and similar workloads.
- New Tyan platforms in its HPC-focused Transport HX product line and database-focused Transport SX product line are also powered by second-gen Epyc processors.
- Additionally, Gigabyte has four new G-Series GPU servers that support the processor.
As for end customers, AMD is announcing its re-entry onto the Top500 list of the world's fastest supercomputers with a system from the French high-performance computing organization, GENCI. The organization announced the latest extension of the Joliot-Curie supercomputer based on the Atos BullSequana XH2000, using the EPYC 7H12 processor.
Other customer wins include the San Diego Supercomputer Center's new Expanse supercomputer system, powered by Dell EMC PowerEdge servers and second-gen Epyc processors. Expanse, has 728 dual-socket Dell EMC PowerEdge C6525 with 2nd generation AMD Epyc processors. Expanse has a projected peak performance of up to five petaflops.
ETH Zurich (the Swiss Federal Institute of Technology)will power the new Euler VI supercomputer with Epyc 7742 processors. The UK Atomic Weapons Establishment's new Vulcan supercomputer is powered by second-gen Epyc processors, as is the UK's fastest new supercomputer, the ARCHER2.
Meanwhile, AMD is also introducing version 3.0 of ROCm, its open source, programming language-independent platform for GPU computing. It now supports HIP-clang, a compiler built upon LLVM, and offers improved CUDA conversion capability. It has expanded acceleration support for HPC programing models and applications, like OpenMP programing, LAMMPS and NAMD. There's also new support for system and workload deployment tools like Kubernetes, Singularity, SLURM, TAU and others. There's also now ROCm upstream integration into leading TensorFlow and PyTorch machine learning frameworks for applications like reinforcement learning, autonomous driving, and image and video detection.