NVIDIA’s Cambridge-1, Their EPYC New Supercomputer

Most Powerful Supercomputer In Essex … And The Entire UK

For the design of the Cambridge-1 supercomputer, NVIDIA went with fast and good as opposed to cheap, rolling out the entire build in less than 20 weeks but coming in about £32 million over budget.   The supercomputer consists of 80 of NVIDIA’s DGX A100 systems which are connected via Infiniband, with each DGX A100 consisting of eight A100 GPUs, which totals up to 4,423,680 CUDA cores and 320GB of GPU RAM.  Each DGX A100 also contains a pair of AMD EPYC 7742 CPUs, as well as up to 2TB system RAM, 30TB of NVME data cache drives, and two 1.92TB NVME SSDs for storage.

In order to conform with the EU’s environmental standards, NVIDIA uses 100% renewable energy to power the Cambridge-1 and 100% free air cooling; likely to the dismay of any Brexiteers that are aware of this.  The supercomputer was designed specifically for medical research and analysis, The Register spoke with a representative of KCL, a company which specializes in the automated analysis of MRI scans, and who will be making use of the AI capabilities of Cambridge-1 for free.  Indeed, NVIDIA themselves provided £40 million of the total £72 million total cost.  The speculation is that NVIDIA might be on a PR campaign to convince any interested parties that their purchase of ARM will be good for the UK.

NVIDIA is working on a new supercomputer in which they intend to replace the AMD processors with ARM based ones, part of their promise to turn the ARM campus in Cambridge into a world-class AI research centre.  It will be interesting to see what effect, if any, this has on UK legislators.