Sign in / up

HPE Apollo 80 – Advantages and Experiences on Isambard 2 Workloads

 

Published February 2021
apollo_80

HPE’s Andy Warner and Simon McIntosh-Smith, of University of Bristol discuss Apollo 80 and Fujitsu’s A64FX as it is deployed for workloads on Isambard 2, the largest Arm-based supercomputer in Europe. In this discussion:

  • Bringing the benefits of High-Bandwidth Memory (HBM) from GPU to CPU w/wide vector units to mainstream scientific MPI-based codes
  • Rebalancing Memory architecture
    • HBM delivers vast memory bandwidth vs. DIMMs
    • HPC Memory utilization study insights - UK National Supercomputer
    • Overprovisioning of memory
    • Rebalancing with HBM - efficiency and usability
  • Application porting and experiences using Isambard 2