Main similarities/differences between Cirrus NCR and ARCHER2¶
This section provides an overview of the main differences between the ARCHER2 system and Cirrus NCR along with links to more information where appropriate.
Similarities¶
- Cirrus NCR is an HPE Cray EX system like ARCHER2
- Cirrus NCR provides the HPE Cray Programming Environment (CPE) in a similar way to ARCHER2 so many of the same libraries, tools and compilers are available (though they are more recent versions on Cirrus)
Hardware¶
- Cirrus is a much smaller system - there are only 640 compute nodes on Cirrus (compared to 5860 on ARCHER2)
- There are no GPU nodes on Cirrus NCR
- There are 288 cores on a Cirrus NCR compute node rather than 128 on ARCHER2
- Cirrus NCR uses the HPE Cray Slingshot 11 interconnect rather than the Slingshot 10 interconnect used on ARCHER2
- Cirrus NCR compute nodes can access the internet directly
- There are no separate "serial" nodes
- For more information see the Hardware section in the User Guide
Software¶
- The software provided by the Cirrus support team is delivered via Spack rather than manual install processes used on ARCHER2
- We will typically advise that users use Spack to build their own software on Cirrus NCR wherever possible
- A smaller number of research software packages are provided centrally on Cirrus compared to ARCHER2
Slurm scheduler configuration¶
- Many of the partitions and QoS are similar between the two systems. See the Scheduler secion of the User Guide