Fractional GPU support

Nomad
1

Hi, We would like to know whether NOMAD offers Fractional GPU support with NVIDIA GPU Device Plugin. For example, I have a nomad client instance with 1 GPU of 16GB memory(GPU1). I have two jobs each requires 8GB of memory to successfully complete. So I am wondering is there a way to allocate my two jobs to run parallel on the same GPU(GPU1) which means Job1 should take 8GB of GPU1 and remaining 8GB of GPU1 should be used for Job2

1 Like
2

Four years later Iā€™m wondering if this is supported now? It would be a really helpful if the number of GPU cores can be configured as floats or based on VRAM requirements.

The nvidia gpu device plugin seems to support MIG, but that forces one to split the GPU into separated virtual GPUs.

3

I am looking for the same. Cant allocate a full GPU for a service that requires too little VRAM.