Overview of specialty machines

The specialty machines controlled by the Slurm workload manager include:

• head node
• CPU clusters
  • hpc-pr3
  • hagrid and hagrid-storage -- for members of the Bauch Lab only
• GPU servers
  • gpu-pr1-01
  • gpu-pr1-02
  • gpu-pr1-03 request access and include summary of research
  • gpu-pr1-04

Head node

• hostname rsubmit.math.private.uwaterloo.ca
• head node is for job submission and short tasks only, e.g. code compilation, not for executing jobs

CPU clusters

This resource is intended for medium-sized parallel jobs (OpenMP or MPI) and for developing and testing parallel code. Large jobs are better served by the Digital Research Alliance of Canada (formerly Compute Canada).

• hpc-pr3 cluster

  compute node hardware specifications

  #Nodes	8
  Node names	hpc-pr3-01 to hpc-pr3-08
  CPU model (2 per node)	Xeon(R) Gold 6326 2.9 GHz (Ice Lake)
  #Cores per node	32
  Threads per core	2
  System Memory per node	128 GB

The Hagrid cluster is for members of the Bauch Lab.

• hagrid cluster

  hardware specifications 

  #Nodes	8 compute, 1 storage
  Node names	hagrid01 to hagrid08, hagrid-storage
  CPU model (2 per node)	Xeon(R) Silver 4114 CPU @ 2.20GHz
  #Cores per node	10
  Threads per core	1
  System Memory per node	187 GB

GPU servers

This resource is intended for GPU-specific computing on a small scale. Medium and large-scale GPU jobs are better served by SHARCNET.

• research GPU server hardware specifications

  Node name	gpu-pr1-01	gpu-pr1-02	gpu-pr1-03	gpu-pr1-04
  CPU model	2x Intel Xeon E5-2680v4 2.4 GHz (Broadwell)	2x AMD EPYC 7542 2.9 GHz	2x Intel Xeon 8480+ 2.0 GHz (Sapphire Rapids)	1x AMD EPYC 9634 2.25 GHz
  # Cores	28	64	56	84
  Threads per core	1	1	1	1
  System Memory	128 GB	1024 GB	1024 GB	768 GB
  GPU Type	NVIDIA Tesla P100	NVIDIA Ampere A100 PCI	NVIDIA Hopper H100 SXM	NVIDIA L40S
  # of GPU devices	4	8	4	3
  GPU memory per device	16 GB	four with 40 GB, four with 80 GB	80 GB	48 GB