SiPearl, a French firm that designs Arm-based processors for supercomputers, on Monday teamed up with AMD to offer a joint providing for exascale supercomputers. Below the phrases of the deal, SiPearl will present its Rhea processors, whereas AMD will supply its Intuition compute GPUs.
The collaboration between the 2 firms will embody constructing software program that might be suitable with Arm-based Rhea processors, in addition to “assess[ing] the interoperability” of AMD’s ROCm open software program with the SiPearl Rhea CPU. AMD’s ROCm is an open platform close-to-metal software program stack for GPUs. Apparently, SiPearl wish to use it to program its CPUs as properly. Understand that HPC software program for Arm is in its infancy. So maybe making AMD’s ROCm and applications that use it suitable with Rhea is sensible from a time-to-market perspective.
SiPearl’s Rhea system-on-chip (SoC) for high-performance computing (HPC) purposes packs 72 Arm Neoverse Zeus cores, interconnected utilizing a mesh community. It additionally includes a distinctive reminiscence subsystem comprising of 4 HBM2E reminiscence stacks and 4 or six common DDR5 reminiscence channels. With a reminiscence subsystem like this, Rhea will have the ability to profit from the intense bandwidth enabled by HBM2E reminiscence, in addition to the excessive reminiscence capability enabled by DDR5.
Whereas Rhea’s CPU guarantees to supply fairly severe efficiency, it doesn’t appear like that alone will have the ability to energy a supercomputer with efficiency of over one FP64 ExaFLOPS, no less than not with affordable energy consumption. Subsequently, it is sensible to pair the CPUs with HPC accelerators or compute GPUs like AMD’s Intuition MI250X or its successors. SiPearl additionally implies that AMD’s compute GPUs is not going to be the one selection for supercomputer operators adopting Rhea.
“We welcome this collaboration with AMD that drive innovation within the HPC market for a few years,” mentioned Philippe Notton, SiPearl’s CEO and founder. “This new collaboration with a world chief will additional enrich our joint providing of our HPC microprocessor mixed with companion acceleration options. Offering a broader selection for European supercomputers end-users, it can allow Europe to sort out the nice challenges of our time similar to synthetic intelligence, local weather modeling and medical analysis.”
SiPearl’s Rhea processor is at the moment being applied by Open-Silicon Analysis, a contract developer of chips from India. The CPU might be made by TSMC on its N6 (6nm-class) fabrication course of in 2023.
“Excessive Efficiency Computing is on the coronary heart of AMD; our AMD Intuition accelerators energy the primary supercomputer to move the exaflop barrier and we proceed to help quite a few HPC installations all over the world with our merchandise,”mentioned Brad McCredie, company vp, Knowledge Middle and Accelerated Processing at AMD. “Because the world continues to want extra compute efficiency to drive the following discoveries that may change our society, AMD is delighted to have interaction one step additional within the European Ecosystem with SiPearl to collectively supply a transparent path to satisfy exascale supercomputing calls for in Europe.”