Enabling Multiple Tenants on PyTorch

Since it is common to have multiple users sharing a Intel® Gaudi® AI accelerator server or cluster, Intel Gaudi supports multi-tenancy in the container runtime ecosystem for PyTorch. Multiple users may each run a single process workload on one or more Gaudi processors on the same node simultaneously.

Note

  • Running only one process per device is supported.

  • Partitioning and sharing one device across multiple users is not supported.