Slurm is Hazel's production job scheduler as of 17 August 2026. LSF is retired on 16–17 December 2026, when the last nodes move across.

Where Things Stand

WhenWhat
17 August 2026 Slurm became the production scheduler. New work should be submitted with sbatch. All documentation under Running Jobs describes Slurm.
Now – December Remaining nodes are moved from LSF to Slurm in batches. Both schedulers run during this period, but only one of them runs jobs on any given node — a node is never scheduled by both at once.
16–17 December 2026 The last nodes move and LSF is retired. After this date bsub, bjobs and the rest of the LSF commands stop working, and the LSF documentation is withdrawn.

What Is Still on LSF

170 nodes remain under LSF as of 22 August 2026 — 150 CPU-only nodes (7,040 cores) and 20 GPU nodes. These move to Slurm in batches between now and the December cutover, so this list only ever shrinks. Everything not listed here is already on Slurm.

CPU nodes still on LSF

Hosts Nodes CPU Cores/node Memory/node
c001n[01-04]c019n[01-04]76Xeon Gold 6226 (Cascade Lake)32188 GB
c056n[01-04]4Xeon Gold 6130 (Skylake)32188 GB
c057, c058, c066, c068 n[01-04]16Xeon Platinum 8358 (Ice Lake)64251 GB
c063, c064, c065, c067 n[01-04]16Xeon Platinum 8358 (Ice Lake)64503 GB
n0353n038836Xeon Platinum 8462 (Sapphire Rapids)64503 GB
n0391, n03922Xeon Platinum 8462 (Sapphire Rapids)641 TB
150 nodes, 7,040 cores

GPU nodes still on LSF

GPU Nodes Hosts
L40S13gpu19gpu25, gpu27gpu31, gpu34
H1003gpu16, gpu32, gpu33
H2002gpu35, gpu36
A1001gpu26
L401gpu15
20 nodes

GPU hosts are 64-core Xeon Platinum 8358 with 503 GB, except gpu33 (1 TB) and gpu35gpu36 (EPYC 9335, 1.4 TB).

This table is a snapshot and is not updated automatically. For what is available on Slurm right now, use the commands under Checking Current Resources — those read the live cluster. If you need to know whether a specific host has moved yet, ask HPC support.

If You Are Still Using LSF

Nothing is being taken away without notice, but there is no reason to wait: the nodes above move on a schedule, and converting your scripts once is easier than doing it under a deadline in December.

Checking Current Resources

These read the live cluster, so they are always correct — prefer them to any list on a web page:

si                            # node availability by partition and architecture
si --nodes --all              # every node, including down ones
si --nodes --gpus             # GPU availability
si --nodes --partition compute   # detail for one partition
si --qos                      # the QOS you can submit with, and their limits

sinfo and squeue work too; si and sqos are local wrappers that present the same information more readably. See Monitoring jobs and cluster status.

Why si --nodes --all shows nodes as down. The nodes still on LSF are already configured in Slurm so that they join automatically the moment they are moved. Until then Slurm lists them but reports them down, and no job can be scheduled onto them. A down node in that list is expected, not a fault — it is one of the nodes above waiting its turn.

Reporting a Problem

Contact HPC support. For a job-specific issue, include:

  • Job ID
  • The batch script you submitted
  • Error messages or unexpected behaviour
  • Output from squeue -j JOBID or sacct -j JOBID

Related Pages