Maverick: New queues to support long gpu runs

Posted by Chris Hempel on Nov 27, 2017 4:34:48 PM

Updated on Dec 1, 2017 8:59:34 AM

The runtime limit on the gpu queue has been increased to 24 hours.

Original Posting

Two new queues have been configured on Maverick to accommodate GPU jobs that require more runtime than allowed in the gpu queue. These two queues are configured as follows:

gpu-long
- up to 72 hours runtime, one node per job (i.e. sbatch -N 1 and/or -n 20 or less)
- maximum of 8 jobs allowed in queue per user

gpu-verylong
- up to 120 hours runtime, one node per job
- maximum of 3 jobs allowed in queue per user

These queues are available immediately for use and do not require special permission to access. The gpu queue remains with a 12-hour runtime limit.

Please submit any questions you have via the TACC Consulting System.

https://portal.tacc.utexas.edu/tacc-consulting