Skip to content

Kelvin2 Overview

The NI-HPC centre hosts the Kelvin-2 research cluster. Kelvin2 runs on a Linux (Centos 7) Operating system. Below we will list the resources available.

Compute

  • 96 x 128 core Dell PowerEdge R6525 compute nodes with AMD EPYC 7702 dual 64-Core Processors (786GB RAM).
  • 8 High memory nodes (2TB RAM).
  • 32 x NVIDIA Tesla v100 GPUs in 8 nodes.
  • 16 x NVIDIA Tesla A100 GPUs in 4 nodes.
  • 2PB of lustre parallel file system for scratch storage.

Storage

There are 4 pools of storage on Kelvin2 :

Home

/users/"username"
  • Default place that users login to.
  • 50GB/100K file quota.
  • No automated file deletion - therefore good to store smaller/compressed longer term data.

Scratch

/mnt/scratch2/users/"username"
  • Large, shared storage area
  • No Quota
  • Temporary data solution - once per month a purge will delete any data that hasnt been accessed in the previous 3 months.
  • Mainly used for storing data that is neccassary to running jobs and storing their output.
  • Once results are output, permanent data should be moved to longer term storage ( Home/McClayRDS ).

Temp

/tmp
  • Local scratch disk on nodes.
  • Data will be automatically deleted when session ends.

McClayRDS

/mnt/autofs/mcclayrds-projects
  • QUB users only
  • Available only from login and data movement nodes, not compute nodes.
  • Group quotas enabled.
  • Replicated to secondary site.
  • To apply for storage on McClayRDS, contact the Research Data Management team here.