HPE bags contracts to build HPC beasts for UK and Japan boffins

HPE has tucked newly signed HPC contracts under its belt, including one to focus on AI-driven scientific discoveries for the Tokyo Institute of Technology and another for UK medical and scientific research that will be based on Nvidia's Grace "Superchip" processors.

TSUBAME4.0 is the name of the Japanese system, which is expected to be fully operational by the spring of 2024, according to HPE. The supercomputer will be deployed in a newly constructed facility at the Suzukakedai campus of the Tokyo Institute of Technology (Tokyo Tech).

This supercomputer will used be to handle modeling and simulation workloads for scientific research projects, particularly those involving medicine, materials science, and climate research.

The Japanese system will pack in 240 nodes of HPE's Cray XD6500 hardware, which is actually a rebrand of the Apollo 6500 systems that the company announced last year. Each node has a pair of AMD's 4th generation Epyc processors with 768GB of memory and four of Nvidia's H100 Tensor Core GPUs. It will also use Nvidia's Quantum-2 InfiniBand kit for 400Gbps connectivity between nodes.

This configuration is expected to deliver a theoretical peak performance of 66.8 petaflops at 64-bit double precision, with HPE claiming that this represents 20 times more compute performance than its predecessor, TSUBAME3.0.

The TSUBAME4.0 configuration will be similar enough to the existing TSUBAME hardware deployed at Tokyo Tech to allow for the continued use of existing software assets as well as the introduction of newer code for computational science and technology, HPE said.

"TSUBAME has been supporting our research on cyclic peptide drug discovery, which is anticipated to become the next-generation medicine," Tokyo Tech's School of Computing Professor Yutaka Akiyama said in a statement. He added: "With the significantly accelerated TSUBAME4.0, we look forward to its support in realizing intelligent drug discovery through large-scale molecular simulation and fusing it with deep learning technology in generating predictive models."

Just for the Brunel of it

Nvidia made the announcement about the UK system to coincide with this week's ISC supercomputing conference, as it features the company’s Arm-based Grace Superchip CPU technology, although it is to be built by HPE.

Named Isambard 3, the system will have its own dedicated facility at the Bristol & Bath Science Park and is also expected to come online sometime in the northern hemisphere spring of 2024. The project is being led by the University of Bristol, along with the universities of Bath, Cardiff and Exeter, as part of the GW4 Alliance research consortium.

Isambard 3 follows on from the earlier Isambard 1 and 2 projects, which are also Arm-based supercomputers, but the Grace chip is the first Arm-based server processor from a top-tier silicon vendor optimized specifically for high performance computing, says the proposal for the system notes.

The system will feature 384 of the Grace chips, each of which includes 144 cores, for a grand total of 55,296 cores that are expected to deliver 2.7 petaflops of double precision 64-bit floating point performance while consuming less than 270 kilowatts of power, Nvidia claimed.

Simon McIntosh-Smith, professor of HPC at the University of Bristol and principal investigator for the Isambard project, said in a statement: "The Arm-based NVidia Grace CPU enables the breakthrough energy efficiency required to push the boundaries of scientific discovery and solve some of humanity’s most difficult challenges."

These challenges include AI, life sciences, medical, astrophysics and biotech, but Nvidia claimed the new supercomputer will also be able to create detailed models of complex structures such as wind farms and nuclear fusion reactors, to help researchers working to advance in clean energy technology.

The proposal form lists the estimated value of the project at £1.5 million ($1.87 million), and also discloses that Isambard 3 will provide a Multi Architecture Comparison System, or MACS, which will incorporate 2-4 nodes of "every important new CPU and GPU architecture" to be released during the projected four-year lifetime of the system.

• US supers maintain grip on Top500 list as China seemingly hides its powers
• A lone Nvidia GPU speeds past the physics-straining might of a quantum computer – in these apps at least
• OVH punts hybrid water and immersion cooling for high performance systems
• UK Prime Minister wants £800M to spend on big British iron
• IBM says it's been running 'AI supercomputer' since May but chose now to tell the world

This will enable rigorous architectural comparisons between Arm chips and the latest CPUs and GPUs from Intel, AMD, NVIDIA, and other emerging providers, according to Bristol Uni.

These two projects are the latest in a string of supercomputer wins for HPE, which must be patting itself on the back for buying HPC biz Cray back in 2019. Cray hardware powers many of HPE's recent supercomputer projects, such as the Frontier exascale supercomputer in the US and the European LUMI supercomputer in Finland, for example.

In its most recently reported results for the quarter ended January 31, HPE said revenue was up 12 percent year-on-year to $7.8 billion, claiming this was the highest turnover it had reached since highest since the beginning of 2016. ®