Lenovo Bags HPC Contracts For A Pair Of European Customers

Lenovo has pulled a couple of European supercomputer wins out of the bag, one using Intel chips for Imperial College London and an AMD-based system for the European Institute of Oncology.

The hardware maker, which recently found itself caught out by US President Donald Trump's ever-changing tariff decisions, was chosen for a pair of new high-performance compute (HPC) deployments – an interesting departure from the Cray EX hardware from HPE that has dominated supercomputers in recent years.

Imperial College is set to get it hands on a system called HX2 this year that will be built from ThinkSystem SC750 V4 Neptune servers fitted with Xeon 6 CPUs and direct liquid cooling, as part of its existing ICICLE partnership with Intel and Lenovo.

Once complete, the beast will comprise 192 nodes, each fitted with dual 72-core Xeon 6960P processors, Nvidia H200 GPUs, and using 200 Gigabit Ethernet as its cluster interconnect. HX2 will also use a Weka-based shared storage layer with 2 PB of all-flash storage.

We asked how much memory the system will be configured with, and will update if we get an answer.

The new supercomputer, which is to be deployed into a commercial co-location site, is intended to help Imperial's academics and students accelerate their research and position themselves for careers in HPC/AI, the uni claimed.

"The Imperial investment of £10 million ($13.5 million) for HX2 has ensured that we can provide, in collaboration with Intel and Lenovo, a high-quality, future-proof compute platform for our researchers," said the college's director of Research Computing Services (RCS), Andrew Richards.

Lenovo is also building an HPC deployment for the European Institute of Oncology and the Monzino Cardiology Center in Milan, Italy. Researchers hope to use this to create predictive, prognostic or diagnostic computational models based on the interactions of protein structures, using data available in the clinical data lakes held by the two Institutes.

The unnamed super will comprise ThinkSystem SR645 V3 and ThinkSystem SR685a V3 servers, both of which support AMD's 5th Gen Epyc 9005 processors, fitted with Nvidia H200 GPUs and backed by a ThinkSystem DE6400F all-flash storage system.

• Please tell us Reg: Why are AI PC sales slower than expected?
• Nvidia ain't done with x86 as it taps Intel Xeons to babysit GPUs
• Hyperconverged infrastructure is so hot right now it needs liquid cooling
• Nvidia joins made-in-America party, hopes to flog $500B in homegrown AI supers by 2029

There are nine SR645 V3 computational nodes, two management nodes, two storage nodes, three display nodes, one 8x GPU H200 SR685a V3 node, and one DE6400F as backend storage. The storage capacity is about 100 TiB or 110 TB.

According to Lenovo, as well as designing the HPC infrastructure for the Monzino researchers, it had a hand in the machine learning models that will be used.

"For years, the group has adopted a data-driven strategy for research, care and governance of the institutes," said Chief Information Manager and Chief Data Officer Annarosa Farina. "Our studies, from basic research to advanced clinical research, are based on big data. Lenovo's HPC infrastructure accelerates and streamlines data processing and is critical to implementing our strategy."

Lenovo claimed strong growth for its AI server business in its recently reported earnings for FY24/25, saying its Infrastructure Solutions Group (ISG) revenue was up 63 percent year-on-year to $14.5 billion. ®