What Do I Do If the Driver and Image I Have Selected Do Not Match My Service Scenario?

You have selected a public image with the Tesla driver (or a GRID driver) for your rendering (inference) service. An error occurred when the software was running.
For example, if you select an improper public image for the rendering scenario, the error message "A D3D11-compatible GPU (Feature Level 11.0,Shader Model 5.0) is required to run the engine" is displayed when the software is running.
You have selected a public image without a driver for your rendering (inference) service, and a driver has not been installed independently. Consequently, the GPU is unavailable, and the error message "command not found" is displayed when you run the nvidia-smi command.

Confirm your service scenario.
Check whether the image has a driver that is suitable for the scenario installed.
- If you use a heterogeneous public image, distinguish the driver type and version by image name. If the image name contains with tesla, the Tesla driver is installed for the selected image. If the image name contains with grid, the GRID driver (excluding the license) is installed for the selected image.
- If you use a private image or other images, run the nvidia-smi command to check whether a driver is installed and check the driver type and version.
- If you choose to install the Tesla driver by yourself, ensure that the Tesla driver version matches the CUDA software version. For details, see Obtaining a Tesla Driver and CUDA Toolkit.

If you have not installed the driver, install the driver, switch to a public image that has a driver installed, or use the script to automatically install the driver as follows:
If you have installed the driver but the driver does not match your service scenario, uninstall the driver and reinstall a new one. For details, see GPU Driver.

Parent topic: Self-recovery from Non-hardware Faults