Aoraki Cluster Job Limits

This document describes the job submission and resource limits enforced on the Aoraki cluster. These limits are designed to ensure fair access, system stability, and predictable performance for all users.

Per-User Job Limits

The following limits apply per user, regardless of partition unless otherwise noted:

  • A maximum of 5,000 submitted jobs may be in the queue at any time.

  • Up to 10 OnDemand jobs may be run concurrently and do not count toward the 5,000 job limit.

  • Users are limited to 2 simultaneously running GPU jobs per GPU partition. Any additional GPU jobs will remain queued until resources become available.

Per-GPU Job Limits

Each job requesting GPU resources (including jobs launched via OnDemand) is limited to:

  • Maximum GPUs: 2 GPUs

  • Maximum CPUs: 16 CPUs - RTX3090 and L40_24GB partitions: 8 CPUs

  • Maximum system memory: 150 GB - RTX3090 and L40_24GB partitions: 60 GB

  • Node usage: Single-node only

Per-CPU-Only Job Limits

For CPU-only jobs (including OnDemand jobs):

  • Each job is limited to execution on a single node.

  • The number of simultaneous jobs, CPUs, and system memory per job are constrained by the partition being used, as outlined below.

Partition-Specific Limits

The following table summarises the limits for CPU-only jobs by partition:

Partition

Max Simultaneous Jobs

Max CPUs per Job

Max System Memory per Job

Aoraki (default)

100

126

1000 GB

Aoraki_bigcpu

50

252

1500 GB

Aoraki_fastcore

50

94

1500 GB

Aoraki_bigmem

10

126

2000 GB

Aoraki_long

25

252

2000 GB

Aoraki_short

250

32

256 GB

Aoraki_small

30

8

32 GB

OnDemand

10

252

2000 GB

Notes

  • Jobs that exceed these limits will remain queued until sufficient resources are available.

  • Limits may be adjusted during maintenance periods or in response to system load.

  • Users with special requirements should contact the Research Computing support team before submitting large or unusual workloads.