HX3 is our GPU machine, designed for AI workloads in teaching and research. The Department of Computing worked with us in proposing this to the University Management Board, who supported it as a teaching and research resource for the whole University.

Specification

  • Compute nodes: 16 GPU Nodes providing 64 Nvidia H200 GPUs
    • 4 x NVidia HGX H200 141GB
    • 2 x Intel Granite Rapids 8562Y+ (32c, 2.8Ghz), 64 cores per node
    • 1,536GB RAM
  • Interconnect: 2 x Nvidia SN5600 800GbE network core; 4 x 200 Gb ethernet to GPU nodes;
  • Storage: 774 TB (usable) of Weka all-flash storage (providing >10TB storage per GPU)
  • Platform: Nvidia Run:ai GPU orchestration and optimisation platform to support teaching and research within AI.