The system supports 2 types of user accounts: namely project-based and individual students with approved UROP

• Project-based accounts
  • Allow to access more computational resources, with allocation granted during project approval
  • Computational resources are shared among all group members of the project
  • Usage accounting for computational resource is implemented, details to be announced later
  • Provide shared storage space for the group
     
• Individual student accounts
  • Computational resources are allocated to each student individually
  • No usage accounting

Resource Request

• Resource request is counted by GPU Resource Unit (GRU). Each GRU associated with different maximum CPU cores and system memory in slurm partitions.
   
• For the project & large-project partitions, 1 GRU corresponds to
  • One H800 GPU with 80GB GPU memory
  • 14 CPU cores with 28 Threads
  • 224GB system memory

• For the student partition, each H800 GPU is partitioned into different size of GPU instances using Nvidia MIG technology, with 1 GRU corresponds to either 3g.40gb, 4g.40gb or 7g.80gb MIG device
  • For 3g.40gb, 1 GRU is
    • 3/7 of one H800 GPU system computational power with 40GB GPU memory
  • For 4g.40gb, 1 GRU is
    • 4/7 of one H800 GPU system computational power with 40GB GPU memory
  • For 7g.40gb, 1 GRU is
    • equivalent to whole H800 GPU of computational power and memory
  • 8 CPU cores with 16 Threads
  • 160GB system memory
     

• For the debug partition, 1 GRU corresponds to

  • One H800 GPU with 80GB GPU memory

  • 14 CPU cores with 28 Threads

  • 224GB system memory

Partition Table