//Discover SCU17 upgrade and Haswell node decommission

As part of the regular upgrades to the Discover cluster, the next scaleable compute unit (SCU17) has been purchased. The hardware has been delivered and vendor acceptance tests have been completed. The NCCS is now working on integrating SCU17 into the Discover cluster and expects it to be available for general use later this summer. SCU17 consists of 704 nodes of AMD Milan processors, with 128 CPUs per node.

As part of the integration process, the remaining Haswell compute nodes on Discover will need to be decommissioned over the next 2 to 3 weeks. The exact date depends on work related to the integration of SCU17 into the Discover cluster. We will send an additional notice a few days prior to the final decommission.

If you are still requesting Haswell nodes via: #SBATCH - -constraint=hasw, those jobs will be rejected once the Haswell nodes are removed. Please contact the NCCS User Support at support@nccs.nasa.gov if you need assistance ensuring that your work can run on the Skylake and/or Cascade Lake processors on Discover.

We recognize this will cause a further reduction in compute capacity on Discover and we are working hard to make the SCU17 integration happen as quickly as possible to limit the impact to your work.