SiPearl, a French company that designs Arm-based processors for supercomputers, on Monday teamed up with AMD to provide a joint offering for exascale supercomputers. Under the terms of the deal, SiPearl will provide its Rhea processors, while AMD will offer its Instinct compute GPUs.
The collaboration between the two companies will include building software that will be compatible with Arm-based Rhea processors, as well as "assess[ing] the interoperability" of AMD's ROCm open software with the SiPearl Rhea CPU. AMD's ROCm is an open platform close-to-metal software stack for GPUs. Apparently, SiPearl would like to use it to program its CPUs as well. Keep in mind that HPC software for Arm is in its infancy. So perhaps making AMD's ROCm and programs that use it compatible with Rhea makes sense from a time-to-market perspective.
SiPearl's Rhea system-on-chip (SoC) for high-performance computing (HPC) applications packs 72 Arm Neoverse Zeus cores, interconnected using a mesh network. It also features a unique memory subsystem comprising of four HBM2E memory stacks and four or six regular DDR5 memory channels. With a memory subsystem like this, Rhea will be able to benefit from the extreme bandwidth enabled by HBM2E memory, as well as the high memory capacity enabled by DDR5.
While Rhea's CPU promises to offer quite serious performance, it does not look like that alone will be able to power a supercomputer with performance of over one FP64 ExaFLOPS, at least not with reasonable power consumption. Therefore, it makes sense to pair the CPUs with HPC accelerators or compute GPUs like AMD's Instinct MI250X or its successors. SiPearl also implies that AMD's compute GPUs will not be the only choice for supercomputer operators adopting Rhea.
"We welcome this collaboration with AMD that drive innovation in the HPC market for many years," said Philippe Notton, SiPearl’s CEO and founder. "This new collaboration with a global leader will further enrich our joint offering of our HPC microprocessor combined with partner acceleration solutions. Providing a broader choice for European supercomputers end-users, it will enable Europe to tackle the great challenges of our time such as artificial intelligence, climate modeling and medical research."
SiPearl's Rhea processor is currently being implemented by Open-Silicon Research, a contract developer of chips from India. The CPU will be made by TSMC on its N6 (6nm-class) fabrication process in 2023.
"High Performance Computing is at the heart of AMD; our AMD Instinct accelerators power the first supercomputer to pass the exaflop barrier and we continue to support numerous HPC installations around the world with our products,"said Brad McCredie, corporate vice president, Data Center and Accelerated Processing at AMD. "As the world continues to need more compute performance to drive the next discoveries that will change our society, AMD is delighted to engage one step further in the European Ecosystem with SiPearl to jointly offer a clear path to meet exascale supercomputing demands in Europe."
Stay on the Cutting Edge
Join the experts who read Tom's Hardware for the inside track on enthusiast PC tech news — and have for over 25 years. We'll send breaking news and in-depth reviews of CPUs, GPUs, AI, maker hardware and more straight to your inbox.
Anton Shilov is a Freelance News Writer at Tom’s Hardware US. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
AMD strategy, shot everywhere if hit something profit.Reply
SiPearl's Rhea system-on-chip (SoC) for high-performance computing (HPC) applications packs 72 Arm Neoverse Zeus coresZeus is the codename for the Neoverse V1 cores, like those in Amazon's Graviton 3 that launched about an entire year ago! Had to look that up. I don't know why they even bother with code names.
Anyway, Graviton uses TSMC N5, while the SiPerl Rhea is going to use N6, even while launching probably 1.5 years later.
SiPearl's Rhea processor is currently being implemented by Open-Silicon Research, a contract developer of chips from India.Um, wow. I honestly would've expected them to do the development in Europe, with mostly European workers.
Amdlova said:AMD strategy, shot everywhere if hit something profit.
Old sports proverb: You miss 100% of the shots you don't take.