Teaching GPU cluster overview

The teaching GPU cluster consists of a head node and three GPU compute nodes. The head node is a CPU-only machine while each compute node has multiple GPUs. The cluster is controlled by the Slurm workload manager. To use the compute nodes you must submit jobs via the head node, namely tsubmit.math.private.uwaterloo.ca. Undergraduate students registered in non-CS Math courses may use these GPU servers. CS students should use CS-provided resources.

Head node

Node name slurm-pt2-01.math.private.uwaterloo.ca,
alias tsubmit.math.private.uwaterloo.ca
#Nodes 1
CPU model (2) Intel(R) Xeon(R) Gold 6150 CPU @ 2.70GHz
#Cores/Node 36
Threads per core 2
System Memory/Node 188 GB

GPU compute nodes

Node names CPU model (2 per node) #Cores Threads
per core
System
Memory
GPU type # GPUs GPU memory per/device
gpu-pt1-02,
gpu-pt1-03
Intel(R) Xeon(R)
E5-2650v4 @ 2.20GHz
24 2 256 GB NVIDIA GeForce GTX 1080ti 8 12 GB
gpu-pt1-04 AMD(R) EPYC(R) Milan
7643 @ 2.3 GHz
96 2 1 T NVIDIA RTX A6000 Ada 5 48 GB
gpu-pt1-05 AMD EPYC Turin
9255@ 3.25GHz
24 2 768 GB NVIDIA L40S ADA 3 48 GB