New compute nodes added to Hydra#

../../../_images/hpc-nodes-zen5.jpg

We’re excited to announce the immediate availability of a brand new compute partition in Hydra. This partition is called zen5_mpi and it has 20 nodes with 128 CPU cores each and 768 of RAM memory, which results in a large 6GB of memory per core that matches the zen4 partition.

Moreover, the zen5_mpi partition is particularly well suited for multi-node MPI jobs, as hinted by its name. The 20 nodes are connected with a high-speed InfiniBand NDR network at 200 Gbps. This is double the bandwidth of the skylake_mpi partition.

Detailed hardware information is available in VSCdocHydra Hardware

By the end of May, a second new partition of AMD Zen5 nodes will be added to Hydra. This partition will be called zen5_himem and have 4 nodes with the same hardware specification as the zen5_mpi nodes but with double the memory reaching 1.5TB of memory in total and 12GB/core.

Specifications#

  • CPUs: 5th Generation AMD EPYC 9535 (2x 64-Core per node, 6000 MB per core)

  • Network: 200 GBit/s NDR Infiniband (ideal for multi-node MPI jobs)

  • Operating System: Rocky Linux 9

Software Recommendations#

Software is available for the 2022a, 2023a, and 2024a toolchain generations. We recommend using the 2024a toolchain whenever possible to get the best performance on the new zen5 nodes. It must be noted that the 2022a and 2023a software stacks were built with the older Rocky Linux 8, which can potentially cause incompatibility issues with the Rocky Linux 9 operating system used in these new nodes. Please let us know if you encounter any such issues.

Feel free to contact us at VUB-HPC Support with any comments or questions about the new AMD Zen5 nodes in Hydra.

On this page