On-Prem

HPC

Home-grown Euro chipmaker SiPearl signs deal with HPE, Nvidia

Claims partnerships will drive development and adoption of exascale computing in Europe


European microprocessor designer SiPearl revealed deals with Nvidia and HPE today, saying they would up the development of high-performance compute (HPC) and exascale systems on the continent.

Announced to coincide with the ISC 2022 High Performance conference in Hamburg this week, the agreements see SiPearl working with two big dogs in the HPC market: HPE is the owner of supercomputing pioneer Cray and Nvidia is a leader in GPU acceleration.

With HPE, SiPearl said it is working to jointly develop a supercomputer platform that combines HPE's technology and SiPearl's upcoming Rhea processor. Rhea is an Arm-based chip with RISC-V controllers, planned to appear in next-generation exascale computers.

The partnership will expand heterogeneous computing options for supercomputing, according to SiPearl, and is intended to support and accelerate adoption of exascale systems in Europe.

It appears that the contributions from HPE will include high-performance networking with the Slingshot interconnect, plus a software compiler and programming stack that will be based on the HPE Cray Programming Environment.

Slingshot, which HPE inherited from Cray, is a high-performance network standard that is also compatible with Ethernet and features in many current supercomputers, including the Frontier system that has now claimed first place in the Top500 rankings.

The HPE Cray Programming Environment is a suite of development software comprising compilers, programming languages, scientific, mathematics, and I/O libraries, plus profiling and performance optimization tools.

The partnership is expected to deliver joint HPE and SiPearl solutions to HPE's Center of Excellence (CoE) in Grenoble, which has recently deployed a new AI supercomputer called Champollion.

"In addition to our recent investments in a new HPC factory in Czech Republic and an AI cluster, dedicated to research, in Grenoble, France, our partnership with SiPearl is another demonstration of HPE's ongoing commitment to support Europe's mission to deliver world-class supercomputing that extends its sovereignty and competitiveness on the global stage," HPE executive vice president for HPC & AI Justin Hotard said in a statement.

With Nvidia, SiPearl said it has entered into a "strategic collaboration agreement" for joint technical and business development that will combine their respective hardware and software solutions.

The two said they intend to develop a proxy platform for porting activities that will combine SiPearl's processors and Nvidia's GPUs. The work will also include joint efforts with other European research institutions on elements including simulation capabilities in open-source and research-oriented modeling tools.

Essentially, the partnership is about SiPearl and Nvidia agreeing to ensure that Nvidia's accelerated computing and networking portfolio will operate seamlessly with SiPearl's Rhea processor and future CPUs, as well as working to expand the Arm HPC ecosystem itself.

Nvidia doesn't just do GPUs, of course, as it acquired network specialist Mellanox for its InfiniBand and Ethernet tech in 2020.

Today's agreement between the companies was given the thumbs-up by some in the European HPC sector.

"It holds great potential for the European supercomputing space, the Arm ecosystem and accelerated computing," said Dr Thomas Lippert, head of the Jülich Supercomputing Centre in Germany.

Last year, SiPearl also announced a collaboration with Intel over the latter's Ponte Vecchio GPU technology, making Arm-based supercomputers with Intel GPUs possible. As part of that agreement, SiPearl said it would port and optimize Intel's oneAPI heterogeneous compute toolkit for the Rhea processor.

That announcement also implied that Ponte Vecchio would feature in the first European exascale supercomputer.

SiPearl itself was born out of the European Processor Initiative (EPI), aimed at developing a new generation of high-end European microprocessors.

The Rhea processor is set to be the first fruit of this, and is due to launch sometime before the end of this year. It is expected to comprise 72 of the Neoverse V1 (codenamed Zeus) 64-bit Arm cores, plus support for DDR5 or the high-performance HBM2E stacked memory technology. ®

Send us news
2 Comments

HPE targets enterprises with Nvidia-powered platform for tuning AI

'We feel like enterprises are either going to become AI powered, or they're going to become obsolete'

HPE says impact of AI on enterprise not 'overstated.' It must be hoping so

Company counting on widespread business adoption to counter server declines

HPE and Nvidia offer 'turnkey' supercomputer for AI training

If you can afford it – pricing's not out yet

DoE watchdog warns of poor maintenance at home of Frontier exascale system

Report says new QA plan currently being worked up

Server sales down 31% at HPE as enterprises hack spending

Customers still 'digesting' shipments bought in 2022, says exec in mixed fiscal year for hardware biz

Dell APJ chief: Industry won't wait for Nvidia H100

Canalys mostly agrees, but thinks GPU giant still has a way to go

Nvidia sees Huawei, Intel in rear mirror as it grapples with China ban

Commerce Sec threatens redesigned AI-capable chips will be curtailed 'the very next day'

Nvidia’s China-market H20 chips hit another speed bump

Integration woes delay Nvidia's hopes of maintaining grip on Middle Kingdom

After bashing Nvidia for ‘arming’ China, Cerebras's backer G42 alarms US govt with suspected Beijing ties

What was it they say about folks in silicon houses?

HPE to start pumping AI capabilities into Greenlake under Project Ethan

OpsRamp now natively available through IT-as-a-service platform

Thirty-nine weeks: That's how long you'll be waiting for an AI server from Dell

Revenue and net income down, server market flickers, PCs fail to ignite

Nvidia intros the 'SuperNIC' – it's like a SmartNIC, DPU or IPU, but more super

If you're doing AI but would rather not do InfiniBand, this NIC is for you