asdf-awscli VS aws-virtual-gpu-device-plugin

❯ asdf plugin-list --urls --refs act https://github.com/grimoh/asdf-act.git master 8729029 aws-vault https://github.com/beardix/asdf-aws-vault.git master 937a1db awscli https://github.com/MetricMike/asdf-awscli.git main b9ba4c7 docker-slim https://github.com/everpeace/asdf-docker-slim.git master 4ee75a3 golang https://github.com/kennyp/asdf-golang.git master 1f388f1 helm https://github.com/Antiarchitect/asdf-helm.git master 87eef5a java https://github.com/halcyon/asdf-java.git master f0c702f jq https://github.com/ryodocx/asdf-jq.git master 3144577 kubectl https://github.com/asdf-community/asdf-kubectl.git master da7bb0b minikube https://github.com/alvarobp/asdf-minikube.git master 8ca7b8d mysql https://github.com/iroddis/asdf-mysql.git master 3aaf756 nodejs https://github.com/asdf-vm/asdf-nodejs.git master cb61e3d perl https://github.com/ouest/asdf-perl.git master 31bb799 php https://github.com/asdf-community/asdf-php.git master 759843b postgres https://github.com/smashedtoatoms/asdf-postgres.git master 4f8b356 python https://github.com/danhper/asdf-python.git master 8ab052f redis https://github.com/smashedtoatoms/asdf-redis.git master bf1276e ruby https://github.com/asdf-vm/asdf-ruby.git master f134c2d sbt https://github.com/bram2000/asdf-sbt.git master 33f9637 scala https://github.com/asdf-community/asdf-scala.git master 1206055 skaffold https://github.com/virtualstaticvoid/asdf-skaffold.git master c942ecf spark https://github.com/joshuaballoch/asdf-spark.git master 6fe49de

This project (available here) uses the k8s device plugin described by this AWS blog post to make GPU-based nodes publish the amount of GPU resource they have available. Instead of the amount of VRAM available or some abstract metric, this plugin advertises the amount of pods/processes that can be connected to the GPU. This is controlled by what is called by NVIDIA as Multi-Process Service (MPS).

AWS has apparently already started using this type of tech as of this year (see lost below). They mention virtual gpus but this particular solution probably won't help OP unfortunately. https://aws.amazon.com/blogs/opensource/virtual-gpu-device-plugin-for-inference-workload-in-kubernetes/

The post explores GPU device plugin to address how to set fractional number of GPU resource for each pod by implementing the Kubernetes device plugin and Nvidia MPS. This project has been open sourced on GitHub.