After considering user feedback and performing operational analysis of the p_gpu_devel queue, we are making the following changes to the PBS queue structure for the Cabeus Milan A100 GPU nodes. These changes take effect immediately.

• Removing the non-prime restriction for the queue to allow the new gpu_devel queue to run during both prime and non-prime time.
• Removing the current p_gpu_devel queue and replacing it with the gpu_devel queue.

Updated Notification (August 16)

The new PBS queues for running jobs on the Cabeus Milan A100 nodes are now available. 

Original Notification (August 12)

Starting at 7:00 a.m. (Pacific Time) on Friday, August 16, the PBS queue structure for the Cabeus Milan A100 GPU nodes will be changed to better facilitate work running on these nodes. At that time, three new queues will go into production, and the previous queues will no longer accept new jobs.

Any jobs already queued in the existing queues or running before 7:00 a.m.on Friday will run without any additional action on your part. Any new jobs submitted after that time will need to use one of the new queues.

Note: Cabeus V100 nodes and Rome A100 nodes are not impacted by these changes.

The new queue structure utilizes the PBS primetime feature. A “non-primetime” queue, np_gpu_wide, will run jobs only from 7:00 a.m. Thursday to 7:00 a.m. Saturday (Pacific Time). The "primetime" queues, p_gpu_normal and p_gpu_devel, will run during the rest of the week, and will also run during the non-primetime days if there are no jobs in np_gpu_wide queue. You can submit jobs to the np_gpu_wide queue at any time during the week, but the jobs will run only during this period.

Here are more details about the queues that will go into production on August 16:

• p_gpu_normal (primetime)
  • Maximum wall time: 24 hrs
  • Maximum node count: 32
• p_gpu_devel (primetime)
  • Maximum wall time: 2 hrs
  • Maximum node count: 2
  • Maximum user jobs (queued or running): 1
  • Priority: 20
• np_gpu_wide (non-primetime)
  • Maximum wall time: 12 hrs
  • Maximum node count: 64
  • Minimum node count: 24
  • Maximum user jobs (queued or running): 2
  • Priority: 20

The following queues will be disabled:

• gpu_12h_64n 
• gpu_24h_32n 
• gpu_36h_16n 
• gpu_devel 

For more information about using Cabeus A100 nodes, see the following HECC Knowledge Base articles: 

• Preparing to Run on Cabeus A100 GPU Nodes 
• Requesting GPU Resources 
• PBS Job Queue Structure

If you have any questions or need more information, please contact the Control Room at (800) 331-8737, (650) 604-4444, or support@nas.nasa.gov.