Expanding Betzy and Saga
Author: Kjersti Strømme
After the expansion, Betzy will support GPU accelerated jobs. Betzy will be expanded by 4 GPU nodes and approximately 5.3PB of extra storage. Each GPU node contains 2x AMD Rome 32 core CPUs and 4x NVIDIA A-100 GPUs. Previously, GPU nodes were only available on Saga and the NIRD Service Platform.
The Saga expansion does not add any new functionality to the system but will increase available resources. The expansion adds 120 compute nodes to the system, giving the machine a total of 320 compute nodes. Each new compute node will contain 2x Intel Xeon-G 6230R 26 core CPU and 192 GB RAM. This will result in an approximate 64% increase in compute capacity.
95 nodes on Fram freed
During periods 2020.1 and 2020.2, the availability of resources on Saga and Fram was limited by Covid-19 reservations made by the Norwegian Institute of Public Health (Folkehelseinstituttet – FHI).
Since 15 February, the covid reservation has been for 70 nodes on Saga and 191 nodes on Fram. The high load has been apparent through the long queue times on the machines. From 1 April, the FHI reservation on Fram has been decreased to 96 nodes, while the reservation on Saga is unchanged. Combining this with the expansion described above should mean a significant increase in resource availability on both Fram and Saga.
One of the compute nodes: The four GPUs, placed underneath their respective cold plates, take up impressively little space because of the direct water cooling.