1565523 : booking 12 GPU nodes on Jan 15

Created: 2026-01-08T15:11:15Z - current status: new

Here is the anonymized and summarized version of the query:

Summary of the Issue

A user requests a reservation of 12 GPU nodes on the Solaris subcluster of the Maxwell cluster for a workshop scheduled on [DATE] from 08:00 to 12:00. The user confirms that the Solaris partition is acceptable for their needs and asks if the reservation is still feasible.

Possible Solution/Next Steps

  1. Check Availability:
  2. Verify if 12 GPU nodes are available in the solgpu partition of the Solaris subcluster for the requested time slot.

  3. The Solaris subcluster has 15 GPU nodes (P100), but some may already be reserved or in use (see context below).

  4. Reservation Process:

  5. If resources are available, the reservation can be created using Slurm commands (e.g., scontrol create reservation).
  6. Example command (adjust as needed): bash scontrol create reservation starttime=2026-01-15T08:00:00 duration=4:00 user=[USERNAME] nodes=12 partition=solgpu flags=overlap

  7. Ensure the reservation is exclusive to avoid interference with other users.

  8. Fallback Options:

  9. If 12 nodes are unavailable, consider:

    • Reducing the number of nodes (e.g., 8–10).
    • Splitting the workshop into multiple sessions.
    • Using alternative partitions (e.g., solarm for ARM-based tasks, though not GPU-enabled).

  10. Confirmation:

  11. Once reserved, share the reservation ID and node list with the user for verification.

Key Context Used

  • The Solaris subcluster has 15 GPU nodes (P100) in the solgpu partition (source).
  • GPU nodes support hardware-accelerated graphical applications via VirtualGL (source).
  • Reservations must be requested in advance due to shared resource constraints.

Sources: 1. Solaris Subcluster Documentation – GPU Nodes 2. Slurm Reservation Guide (External) (for reference on commands).