This month we are complete deliver Machine learning env with Opensource scheduler solution. Customer will be use 2 way of ML/DL. Resource management software and Nvidia docker container env. Both of service can be isolate GPGPU resources.
We are deliver AI stack on CentOS7. That is V100 + Slurm + Nvidia MPS design. Interactive job works well with Slurm. That is one of good option for AI environment stack.