Running w/ multiple GPU (HPC)

[sin-2pi]

Hi guys,

I asked this a while ago but I am wondering if anyone has experience running ODM while utilizing multiple GPUs from a singularity container? This is being ran on a cluster.

Or any advice on a direction to move forward to would be very helpful.

[kikislater]

So it’s the same question as this one:

[HPC] Running on multiple GPUs

It appears as if my second GPU is not being utilized when running ODM on my cluster. Does ODM still only support single GPU processing?

Even if singularity could work with any docker image, docker images run as root. So it might work with singularity but it’s a workaround.

As far as the GPU is concerned, it depends on what you want to do:

• If you want to allocate part of the resources (like cpu you allocate X number of cores) for a singularity node then you need to install the nvidia driver. This specific use requires a non-free driver. It’s not really used except in render farms/
• if you want to have more gpu to compute a task, you need to turn to kubernetes and nvidia => GitHub - NVIDIA/k8s-dra-driver: Dynamic Resource Allocation (DRA) for NVIDIA GPUs in Kubernetes

For singularity, you need a sysadmin with solid knowledge of resource allocation. The simplest use is to allocate an entire GPU to a VM.

[sin-2pi]

Thank you!