NVIDIA / dcgm-exporter

NVIDIA GPU metrics exporter for Prometheus leveraging DCGM
Apache License 2.0
923 stars 159 forks source link

How to install dcgm-exporter on Windows Server? #344

Closed LittleNewton closed 5 months ago

LittleNewton commented 5 months ago

Ask your question

Is there a release for Windows Server 2022?

I know almost all GPU servers are Linux-based, but a few are Windows-based.

### Tasks
nvvfedorov commented 5 months ago

Can you run dcgm-exporter in a docker container?

LittleNewton commented 5 months ago

@nvvfedorov Yes! I can run it in Linux docker container. So do you mean dcgm-exporter can work in windows docker desktop app?

nvvfedorov commented 5 months ago

@LittleNewton , We never tried, but it worth to try. According to Docker Desktop documentation, Windows grant access to GPU for Docker containers. See: https://docs.docker.com/desktop/gpu/.

nvvfedorov commented 5 months ago

Also, you may try to use Ubuntu on WSL: https://ubuntu.com/blog/getting-started-with-cuda-on-ubuntu-on-wsl-2.

LittleNewton commented 5 months ago

@nvvfedorov , thanks for your advice. I have installed Docker Desktop on my Windows Server 2022 and dcgm-exporter works smoothly in this environment.

Docker Desktop rely
  --> WSL2 rely
      --> Hyper-V rely
            --> Security Policies for hardware isolation and software integrity (X, conflict with VMware Workstation Pro)

Now I give up VMware Workstation Pro because it said VMware Workstation is not compatible with Device/Credential Guard. Anyhow, it works. Thanks a lot!

LittleNewton commented 5 months ago

What's more, if you know how to make VMware workstation Pro and Docker Desktop compatible with each other, please let me know. :)