[slurm-users] QOS or Priority Access to GPU/GRES?

Jason Simms simmsj at lafayette.edu
Tue Apr 27 13:17:28 UTC 2021

Hello all,

As usual, I have a super basic question, so thank you for your patience. I
want to verify the correct syntax to configure a GPU for priority preempt
access via a QOS, much like we are currently doing for a specified number
of cores. When I have created a QOS in the past, I've so far only been
concerned about priority access to a subset of CPUs, so this is my simple
example command once I've created a new QOS:

sacctmgr modify qos <qosname> set GrpCPUs=24

Looking at other examples (such as
https://help.rc.ufl.edu/doc/Account_and_QOS_limits_under_SLURM), it seems
that GrpTRES is used to define priority access to a GPU (but then, they
also have an entry for cpu as a GrpTRES as well, which I don't, so maybe I
need to add that??). Is it as simple as something like this?:

sacctmgr modify qos <qosname> set GrpTRES=gres/gpu=1

Or, do I *also* need to add the cpu to the GrpTRES?:

sacctmgr modify qos <qosname> set GrpTRES=cpu=24,gres/gpu=1 GrpCPUs=24

The syntax is a little unclear from examples in the official documentation,
and I guess this is also a question about the difference between GrpCPUs
and GrpTRES.

Warmest regards,

*Jason L. Simms, Ph.D., M.P.H.*
Manager of Research and High-Performance Computing
XSEDE Campus Champion
Lafayette College
Information Technology Services
710 Sullivan Rd | Easton, PA 18042
Office: 112 Skillman Library
p: (610) 330-5632
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20210427/15225e19/attachment.htm>

More information about the slurm-users mailing list