1577414 : Accesss to maxgpu partition

Created: 2026-02-17T14:57:02Z - current status: new

Here is the anonymized and summarized version of the query:

Summary of Issue

A postdoctoral researcher affiliated with a beamline (e.g., [BEAMLINE_ID]) has collected a large dataset requiring GPU-intensive ptychography processing. Preliminary processing was done on the psgpu partition, but the workload demands >32 GB GPU memory for optimal results. Testing on the allgpu partition (with constraint=GPUx4) improved performance, but the user seeks stable, non-preemptible access to the maxgpu partition to avoid job interruptions.

Suggested Solution

  1. Request Access to maxgpu Partition:
  2. The maxgpu partition is intended for DESY staff with applications suitable for high-performance computing (e.g., GPU-accelerated simulations, large-memory workloads).

  3. The user should email [maxwell.service@desy.de] with:

    • A brief explanation of the application (ptychography processing).
    • Justification for needing >32 GB GPU memory and non-preemptible access.
    • Confirmation that their work aligns with DESY’s HPC suitability criteria (e.g., efficient multi-GPU usage).

  4. Alternative Options:

  5. If maxgpu access is denied, consider:

    • compgpu partition (new HPC Gateway hardware with H200 GPUs, but limited to 2 concurrent jobs/user and subject to preemption).
    • Beamline-specific partitions (if available) or contacting [fs-ec@desy.de] for Photon Science resources.

  6. Acknowledgements:

  7. If granted access, acknowledge the Maxwell cluster (and HPC Gateway, if using compgpu) in publications.

Sources Used

  1. Maxwell Documentation: Access to Partitions
  2. Maxwell Documentation: Photon Science Resources
  3. Maxwell Blog: New GPU Hardware (2025-09-29)