GM4 Cluster Overview

This section of the documentation provides an overview of how the GM4 cluster is organized.

Partitions

The GM4 cluster consists of 29 4-way GPU nodes and 4 CPU-only nodes for a total of 33 nodes. The nodes are accessible thourgh two partitions, gm4 and gm4-pmext. The gm4-pmext partition includes all 33 nodes and is accessible to PI groups affiliated with the Pritzker School of Molecular Engineering (PME). The gm4 partition contains a subset (23 nodes) of the 33 nodes and is accessible to non-PME GM4 participants.

Partition Name	Users	Nodes	Node List
gm4-pmext	PME users	29 GPU nodes and 4 CPU-only nodes	midway2-[0631-0663]
gm4	non-PME GM4 users	19 GPU nodes and 4 CPU-only nodes	midway2-[0641-0663]

All non-PME users will use the gm4 partition to submit jobs, whereas PME users will use the gm4-pmext partition for job submission.

Slurm Quality of Service (QOS)

The fair-share use of the GM4 resources is managed through the Slurm scheduler Quality of Service (QOS) settings. The quality of service defines the type of resource, GPU or CPU-only, that a job can request and whether the job is a production or debug run. There are three QOS that can be used with both the gm4 and gm4-pmext partitions. The QOS settings are the same for the gm4 and the gm4-pmext partitions. The resource settings for each QOS are defined as follows:

QOS Name: gm4

(default if no QOS specified)

		Per User Settings				Per Account Settings
Max Wall Time	QOS Priority	Max Running Jobs	Max Jobs Submit	Max CPUs	Max GPUs	Max Jobs Submit	Max CPUs	Max GPUs
1-12:00:00	10000	28	28	320	32	48	320	32

QOS Name: gm4-cpu

		Per User Settings				Per Account Settings
Max Wall Time	QOS Priority	Max Running Jobs	Max Jobs Submit	Max CPUs	Max GPUs	Max Jobs Submit	Max CPUs	Max GPUs
1-12:00:00	10000	28	28	320	N/A	56	320	N/A

QOS Name: gm4-debug

		Per User Settings				Per Account Settings
Max Wall Time	QOS Priority	Max Running Jobs	Max Jobs Submit	Max CPUs	Max GPUs	Max Jobs Submit	Max CPUs	Max GPUs
0-30:00:00	100000	1	1	40	4	2	80	8