April 30, 2021 — CSC has opened supercomputer Mahti’s new GPU extension, Mahti-AI, for researchers’ use. There is a growing interest among researchers in GPU computing and the need of GPU capacity has been raised in CSC user surveys.
Mahti-AI consists of 24 GPU nodes with the latest generation Nvidia Ampere A100 GPUs. It significantly boosts the resources available for artificial intelligence research, AI workloads and converged AI and HPC workloads. Popular ML/AI frameworks such as PyTorch and TensorFlow are supported from day one.
This purchase is the final significant update in the National Data Management and Computing Development Program (DL2021).
Mahti-AI offers an access to the latest generation NVIDIA GPU hardware and its new capabilities such as improved tensor cores. The GPU nodes are based on the Atos BullSequana X2415 GPU blade each with
two AMD Epyc Rome 7H12 processor (CPU),
- four NVIDIA Ampere A100 processors (GPU)
- 512 GB memory for CPU and 160 GB for GPU (40 GB to each GPU).
- 3.8 TB of local Nvme storage
- two 200 Gbit/s network connections.