The Worldwide Supercomputer Convention (ISC) kicked off in Hamburg, Germany this week with the discharge of the TOPP500 record of the quickest supercomputers, with a pc named Frontier taking first place.
Deployed on the US Division of Vitality’s Oak Ridge Nationwide Labs, it’s the first exascale machine (1018 floating level operations per second)—an HPE-Cray EX system powered by AMD Epyc CPUs and Intuition MI250 GPUs.
Intel had hoped to win the exascale battle with one other DoE laptop known as Aurora, however AMD beat it to the punch. Frontier additionally beat out opponents from China and Japan that had hoped to win the exascale race.
The success of Frontier isn’t a shock. A smaller model named Crusher was deployed on the Oak Ridge labs the place it has been used to validate the purposes to run on Frontier.
It’s yet one more milestone within the comeback of AMD. Its server processors at the moment are greater than aggressive with Intel’s greatest, and Intuition exhibits it might compete with coprocessors made by Nvidia, which has been its actual nemesis. AMD has steadily gained floor on Intel and the server house, however Nvidia just about owns the present relating to AI and HPC.
AMD shouldn’t be popping the champagne simply but. Nvidia coprocessors will be present in 154 of the TOPP500 supercomputers; solely seven of the supercomputers are utilizing AMD Intuition playing cards.
Cray and AMD have another factor to boast about: Frontier can be on the prime of the Green500 record as essentially the most energy-efficient supercomputer.
Intel updates its HPC roadmap
For a while, Intel has talked about its prime HPC processor, Ponte Vecchio, the info heart model of Intel’s Xe GPU structure. Ponte Vecchio competes towards Nvidia’s Ampere A100 and AMD’s Intuition MI2x0 chips, each of that are business GPUs repurposed and barely altered for HPC and AI workloads.
Now Intel is transferring past that processor and giving a glimpse into what’s coming after it: Rialto Bridge. Due subsequent 12 months, the brand new processor will present round 30% higher efficiency than Ponte Vecchio for purposes, based on Jeff McVeigh, vp and normal supervisor of Intel’s Tremendous Compute Group. (Are you able to say Osborne Impact)?
The successor to Rialto Bridge, Falcon Shores, is way more formidable. It’s a merging of each the Ponte Vecchio GPU and Sapphire Speedy-era of Xeon processors in a single package deal. That’s roughly what Nvidia is doing with its Grace Hopper Superchip, which consists of a 72-core CPU and a Hopper H100 GPU tightly linked by Nvidia’s high-speed NVLink-C2C chip-to-chip interconnect. AMD, your transfer.
McVeigh mentioned Falcon Shores will present 5 instances higher performance-per-watt, reminiscence capability, and reminiscence bandwidth than “present platforms,” which doubtless means Nvidia’s A100 and the AMD Epyc server processor.
“Falcon Shores will deliver collectively x86 and Xe architectures, profiting from next-generation packaging, reminiscence, and I/O applied sciences giving big efficiency and effectivity enhancements for programs computing giant information units and coaching gigantic AI fashions,” he mentioned in a video saying the brand new chips.
Nvidia Strains Up OEMs
Approaching the heels of the much-hyped introductions of its Grace CPU and Hopper GPU, Nvidia introduced that Atos, Dell Applied sciences, GIGABYTE, HPE, Inspur, Lenovo, and Supermicro plan to supply servers constructed with each the Grace CPU and Grace Hopper Superchip. All these new programs will use the lately introduced Grace and Grace Hopper reference designs which offer producers the blueprints to quickly prototype and construct Nvidia-endorsed programs.
It additionally introduced some future programs that might be candidates for the TOPP500 record. Los Alamos Nationwide Laboratory introduced that its next-generation supercomputer, Venado, would be the first system within the US to be powered by Grace CPUs. It will likely be billed by Cray and use a mixture of Grace CPU Superchip nodes and Grace Hopper Superchip nodes. When accomplished, its efficiency is predicted to exceed 10 exaFLOPS.
Nvidia introduced one other HPE/Cray supercomputer, this one known as Alps, situated on the Swiss Nationwide Computing Heart. Alps, constructed with the Grace CPU Superchips and meant for analysis in a spread of fields, is predicted to return on-line in 2023.
Copyright © 2022 IDG Communications, Inc.
