PSC is pleased to announce that the Bridges-2 team just added ten (10) additional nodes to Bridges-2! These HPE Cray 670 nodes, each with eight (8) H100-SXM5-80GB GPUs and 2 TB node memory, are interconnected by a high-performance Infiniband network to the Bridges-2 system, will increase Bridges-2 GPU capacity, and should significantly improve the performance of ML and HPC workloads.
Bridges-2 now has three types of GPU nodes: v100-16, v100-32, and h100-80
New nodes:
- Ten (10) HPE Cray 670 h100-80 nodes, with eight (8) H100-SXM5-80GB GPUs each with 80GB of GPU memory and a total of 2TB RAM per node.
Legacy nodes:
- Nine (9) 2-socket Intel Xeon “Skylake” v100-16 nodes containing eight V100 GPUs without NVLink, each with 16GB of GPU memory. These nodes have 192GB RAM per node.
- 24 2-socket Intel Xeon “Cascade Lake” v100-32 nodes with eight V100 GPUs with NVLink, each with 32GB of GPU memory. These nodes have 512GB RAM per node.
Note the memory difference in both System and GPU memory between these types of nodes.
- If your code requires more than 32GB per GPU, specify the h100-80 nodes – Note: You must explicitly request use of these nodes.
- If your code requires 16-32GB per GPU, specify the v100-32 nodes
- If your code can run on 16BG or less, specify the v100-16 nodes
- If you don’t specify a node type, your job will default to a v100 node.
for an interactive session, use –gres=gpu:type:n
for a batch job, use –gpus=type:n where type is one of v100-16, v100-32 or h100-80.
Should you have any questions or problems, please send email to help@psc.edu and our team will be happy to assist you.
Bridges-2 Team