New GPU power for machine learning and HPC

CSC is expanding its GPU computing capacity to especially meet the rapidly growing needs of the machine learning community. 26 nodes with four Nvidia Pascal P100 GPUs in each node will be available for CSC's customers in September. Accordingly, there are 104 Pascal GPUs in total.

In addition to the four GPUs each node contains two 14 core Intel Xeon E5-2680 v4 2.4 GHz Broadwell CPUs with 512 GB memory. The GPUs are connected via PCIe lanes.

The new nodes are divided between the Taito HPC cluster and the cPouta IaaS cloud platform. Depending on user demands, the exact split can be adjusted relatively easily.

– The role of GPU computing has become central in modern machine learning. Available computing capacity directly defines how large and realistic models the researchers are able to work with. Major GPU capacity is a crucial for modern machine learning research and therefore an important part of CSC's computing service, says Aleksi Kallio, development manager at CSC.

We will notify the users as soon as the new GPU resources are ready for production.