Extra Ampere GPUs in Hydra#

We are happy to announce the immediate availability of 2 more GPU nodes in Hydra adding 4 new Nvidia A100 GPUs to the cluster. These new nodes are joining the ampere_gpu partition and share the same CPU, memory and GPU configuration with the other nodes in this partition.

  • 2x Nvidia A100 GPU (40 GB of GPU memory)

  • 2x AMD Epyc 7282 CPUs (16 CPU cores)

  • 256 GB CPU memory

  • EDR (100Gbit) InfiniBand network

  • 25 Gbit Ethernet network

The total capacity of the ampere_gpu partition at the time of this writing is 18x Nvidia A100.

These 2 nodes have a very fast local scratch storage made of SSD drives with a total capacity of 5.9 TB. This fast SSD storage will increase the performance of jobs that access big databases intensively with random patterns, such as AlphaFold.

Only 4 of the nodes in ampere_gpu have this fast storage though. You can submit your jobs specifically to this type of nodes by using the sbatch option --constraint="big_local_ssd". The fast local scratch in these nodes is accessible as usual, through the environment variables $VSC_SCRATCH_NODE or $TMPDIR. In the future we will provide an option to automatically transfer large databases to the fast local scratch of these new nodes.

A wide range of software is already available on these nodes (module avail), but if you need anything else, please let us know at VUB-HPC Support.