Using NVIDIA DGX Systems¶
Warning
Tier 2 support: NVidia DGX support is experimental and covered by Tier 2 support
You can use containerized execution using DGX Kubernetes cluster as the underlying Kubernetes engine.
Dataiku has been tested for use on DGX Systems and is certified as DGX-Ready Software.
To use DGX Systems you will need to add DGX Kubernetes cluster as an unmanaged cluster in Dataiku.
Additional configuration includes:
Building an image with CUDA support to set up a cluster with a CUDA-enabled base image.
Adding a custom reservation to request multiple GPUs.
Both cloud and on-premises DGX systems have been successfully tested. Multi-Instance GPU (MIG) support has not been tested