GPU
Try it out with
codchi init <NAME> https://github.com/aformatik/codchi nixosModules.gpu.GPU usage should work by default inside a code machine. Codchi uses the GPU driver from your host and therefore only works when the driver is installed on your host. Currently only Nvidia/CUDA is tested.
Nvidia
Check if it works
codchi init gpu_test # create an empty machine
codchi exec gpu_test nvidia-smi
# Output should look similar to this:
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.157 Driver Version: 538.18 CUDA Version: 12.2 |
|-----------------------------------------+----------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 Quadro P500 On | 00000000:02:00.0 Off | N/A |
| N/A 55C P0 N/A / ERR! | 0MiB / 2048MiB | 1% Default |
| | | N/A |
+-----------------------------------------+----------------------+----------------------+
+---------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| No running processes found |
+---------------------------------------------------------------------------------------+
nvidia-container-toolkit
To use your GPU inside Docker inside a code machine, enable codchi.docker.enableNvidia. This also requires requires nixpkgs.config.allowUnfree = true.
For example, here is how you can run InvokeAI inside a code machine:
codchi init invokeai https://github.com/aformatik/codchi nixosModules.gpu
codchi exec invokeai
# on NixOS>=24.11
docker run --runtime=nvidia --device nvidia.com/gpu=all --publish 9090:9090 ghcr.io/invoke-ai/invokeai
# on NixOS<24.11
docker run --runtime=nvidia --gpus=all --publish 9090:9090 ghcr.io/invoke-ai/invokeai
# open web ui in your windows browser (in another terminal)
xdg-open "http://$(ip route | awk '/^default/{print $3; exit}'):9090"
uil:edit Edit this page on GitHub