HPE announces world’s fastest (probably) supercomputer: Discovery at the Oak Ridge National Laboratory

HPE will build a new supercomputer for the US Department of Energy’s Oak Ridge National Laboratory. While it’s careful to avoid saying this will be the world’s fastest supercomputer, Discovery will be the successor to Frontier – which currently takes number two slot behind HPE’s own El Capitan.

In addition, HPE will build “Lux”. It describes this as a “dedicated AI system” and we know it will be built on AMD technology. In particular, it will include AMD Instinct MI355X GPUs, AMD EPYC CPUs and AMD Pensando networking.

โ€œDiscovery will set the stage for a new level of converged HPC, AI and quantum computing capabilities, providing additional insight in connection with other systems, while Lux greatly expands researcher access to dedicated AI resources,” said Bronson Messer, Director of Science for the Oak Ridge Leadership Computing Facility.

He added that the facility expects “unparalleled gains in various, critical areas of scientific research and leadership” due to the new duo.

โ€œWinning the AI race requires new and creative partnerships that will bring together the brightest minds and industries American technology and science has to offer,โ€ said US Secretary of Energy Chris Wright.

Then things got a little political. โ€œThatโ€™s why the Trump administration is announcing the first example of a new commonsense approach to computing partnerships with Lux. We are also announcing, as part of a competitive procurement process, Discovery.

“Working with AMD and HPE, weโ€™re bringing new capacity online faster than ever before, turning shared innovation into national strength, and proving that America leads when private-public partners build together.โ€

HPE Discovery supercomputer: what we know so far

HPE Discovery logo
We don’t have a spec sheet but we do have a logo… (image: HPE)

Discovery will be one of the first supercomputers to be based on the HPE Cray Supercomputing GX5000 platform.

“[This] is the follow on to the GX4000 and it is again purpose built, like our previous generation for supercomputing, and it includes CPUs, GPUs, networking, software, storage and cooling,” said Trish Damkroger, HPE SVP and General Manager, HPC and AI Infrastructure Solutions, during a pre-briefing session.

“It’s completely new architecture, engineered to deliver this unprecedented performance with higher density and truly support these growing workloads and workflows.”

AMD has revealed that the Discovery will use its Instinct MI430X GPUs, while the CPU will use its next-generation EPYC processor codenamed Venice.

โ€œFor more than a decade, AMD and HPE have partnered to push the limits of high-performance computing, delivering solutions that enable discoveries and change the world,โ€ said Dr Lisa Su, CEO of AMD.

โ€œTogether with Oak Ridge National Laboratory, we are advancing the next generation of AI systems with Discovery and Lux – empowering researchers to accelerate innovation and strengthen Americaโ€™s leadership in science and technology.”

HPE says the GX5000 platform will be “augmented” by the new HPE Cray Supercomputing Storage Systems K3000. “It’s built on the DAOS open source [platform] and it’s going to offer you the fastest IOPS right now,” said Damkroger.

The platform will also feature the next generation of HPE’s Slingshot networking infrastructure.

How fast will HPE Discovery be and when will it be available?

HPE is being extremely cagey – for now – about explicit specifications. In terms of performance, all we know for certain about Discovery is that it will again be an exascale computer. And that HPE promises it “will bolster productivity up to 10x”.

Trish Damkroger did drop some hints, however, when talking about the GX5000 platform as a whole. “It’s going to accommodate 127% more compute performance per compute slot. So that’s saying each tray will be able to accommodate these higher TDP parts, and it can accommodate up to 25 kilowatts per compute slot.”

She added: “So not only is going to be more performant, it’s also going to be smaller than previous generations, 42% smaller than the previous generation.”

According to AMD, HPE Discovery is due to be delivered in 2028 with “user operations” starting in 2029. It will almost certainly be the world’s most powerful supercomputer when it comes online.

We expect to hear more, including more details on the specification, at the forthcoming SC25 event, which takes place in St Louis, Missouri, from 16 to 21 November.

Article updated at 12.45pm UK time on 28 October to include details of the AMD technology and the quote from US Secretary of Energy Chris Wright.

Avatar photo
Tim Danton

Tim has worked in IT publishing since the days when all PCs were beige, and is editor-in-chief of the UK's PC Pro magazine. He has been writing about hardware for TechFinitive since 2023.