Atos has signed a four-year contract, £5m (€5.5m) contract with the University of Oxford for a national AI supercomputer built on the Nvidia’s DGX SuperPOD architecture.
The JADE2 system will have 63 DGX nodes, using a total of 504 of Nvidia’s 12nm V100 Tensor Core GPUs, connected with Nvidia Mellanox InfiniBand networking and A3I storage from DataDirect Networks (DDN) in the US. This currently makes the system for the JADE2 project the largest AI supercomputer in the UK but lags behind other European systems.
The largest AI supercomputer system in Europe is coming online at the moment at Forschungszentrum Jülich near Aachen in Germany.
Also built by Atos, this uses a blade with four of Nvidia’s 7nm A100 Tensor Core GPUs and two AMD EPYC CPUs for the BullSequana X2415. Around 120 of the direct water-cooled blades are being added to the existing JUWELS supercomputer to provide 70,000 petaflops of 32bit AI performance.
This follows Nvidia working with University of Florida in the US to build an AI supercomputer with 140 NVIDIA DGX A100 systems powered by 1,120 Nvidia A100 GPUs. It will include 4 petabytes of high-performance storage: Nvidia to build most powerful academic system at Florida
Microsoft is also building a cloud-based commercial AI supercomputer for OpenAI with 285,000 CPU cores and 10,000 GPUs.
In the UK the JADE2 system will be three times the performance of the current JADE1 (Joint Academic Data Science Endeavour, shown above) that works with eight UK universities and the Alan Turing Institute. JADE2 will also be available to an additional 12 UK universities. It will be based at the STFC Hartree Centre in Daresbury, near Warrington.
“The successful delivery of JADE has created more demand among UK researchers and industry for powerful computing facilities which can accommodate high end, data intensive AI workloads,” said Professor Wes Armour at the University of Oxford. “Building on the success of the JADE collaboration with Atos, and by significantly expanding the JADE