Define a Model using GPU
To use GPU for a specific model, please define your model with type cuda-python.
This model type inherites from python, and contains CUDA libraries following requested
environment.
Parameters
| Field | Description |
|---|---|
| cudaVersion | CUDA Veresion to use |
| cudaEnvironment | CUDA environment to use (only Ubuntu environments are supported) |
| pythonVersion | Python version to install |
DeltaTwin GPU instancies DeltaTwin provide GPU instancies with following carateristics:
- CPU: 22
- RAM: 90 Go
- GPU: NVIDIA L4 (24Go GDDR6)
- Storage: 400Go
Only *-ubuntu* environements are supported:
- base-ubuntu22.04
- devel-ubuntu24.04
- runtime-ubuntu26.04
- etc
To retreive the complete list, please check NVIDIA CUDA images.
Caution
Don’t forgot to set the GPU hardware requirement to true, without this the
model will be executed on an instance without GPU.
Example :
{
...
"models": {
"gpu-model": {
"path": "models/gpu-model",
"type": "cuda-python",
"parameters": {
"cudaVersion": "13.0.2",
"cudaEnvironment": "runtime-ubuntu22.04",
"pythonVersion": "3.12",
"hardwareRequirements": {
"cpu": "8000m",
"ram": "4Gi",
"gpu": true
},
...
}
}
}
...
}