Warning
This is as a Lab Notebook which describes how to solve a specific problem at a specific time. Please keep this in mind as you read and use the content. Please pay close attention to the date, version information and other details.
Danger
The amd24 cluster is now in production. This document is obsolete.
Beta Access to the amd24 Cluster
The amd24 cluster is now available for beta access. This means jobs will not run on the new cluster's hardware unless requested. The way to request access depends on whether you want to access a general compute node or a node that belongs to a buy-in.
Please note that the typical restriction on the number of cores in use by a single user at any one time will be increased from 1040 cores to 8256 cores for the beta period.
As this is beta access the nodes may need to restart while we work out any remaining bugs in the system. We will try to avoid interrupting user workloads during this period but some disruption may occur.
dev-amd24 and dev-amd24-h200 are available from the gateway and development nodes via ssh.
Requesting General Compute Nodes
To request a general compute node, users must specify the beta SLURM account. This can be done via the
#SBATCH --account=beta option in SLURM scripts or through the "SLURM account" field under "Advanced Options"
when requesting an OnDemand job. Jobs run on the general nodes have a 24 hour limit during the beta period.
Requesting Buy-In Compute Nodes
To access the new hardware associated with your buy-in, users must specify both their buyin account
(e.g. #SBATCH --account=<buyin> in a SLURM script) and the beta reservation.
In a SLURM script, this combination looks like
1 2 | |
where <buyin> is replaced with your respective buy-in account name.
For OnDemand jobs, the "SLURM Account" and "SLURM Reservation" fields can be found by checking the "Advanced Options" box when specifying app resources.
The normal 1 week queue limit apply to buy-in nodes during beta.