New GPU nodes added to Hydra

We’re happy to announce the public availability of hopper_gpu, a brand new GPU partition in Hydra. This partition features 5 nodes, each equipped with 2 NVIDIA H200 GPUs (140 GB VRAM) and 48 CPU cores.

The H200 GPUs share the same architecture as those in our upcoming Tier-1 supercomputer sofia. This provides an good opportunity to estimate their performance in future Tier-1 project proposals.

We have installed all modules available in 2024a and 2025a toolchains specifically optimized for the hopper_gpu nodes. This brings the collection of available software on this new partition on par to the rest of the cluster. Just keep in mind that software on 2023a toolchains and older might still be loaded on hopper_gpu, but those are not guaranteed to work.

Note

The topology of the hopper_gpu nodes is not yet optimal, which may cause reduced performance in some special cases with high CPU/GPU hybrid loads. We are working with the hardware vendor to resolve this. If you experience such issues in your jobs, please let us know at VUB-HPC Support

Detailed hardware information is available in VSCdocHydra Hardware

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