Determining the Available GPU Drivers

Determining the GPU drivers allows you to confirm the availability of a GPU and driver version for a particular compute. This applies to both gear compute or JupyterLab compute.

Determining GPU Drivers in Gear Compute

1. Look for an installed gear called Flywheel Nvidia CUDA Test on the site. If it does not exist on the site, the gear can be installed from the Flywheel Gear Exchange: Flywheel Nvidia CUDA Test. For installation, use fw beta install

2. Run this gear using the appropriate gear tags to route the gear to a GPU Engine. This gear will output the NVIDIA driver information by running nvidia-smi on the VM.

3. Review the gear log to determine whether the driver is installed. A failed run means no GPU driver and/or GPU. A successful run will provide the GPU name, specifications, driver version, and CUDA compatibility version of the driver.

Determining GPU Drivers in JupyterLab Servers

1. Launch the Notebook using a compute that has GPU support.

2. Open a terminal window using the JupyterLab terminal option.

3. On the command line run nvidia-smi

4. Review the output of the command. If it fails, it means there is no GPU driver and/or GPU. A successful run will provide the GPU name, specifications, driver version, and CUDA compatibility version of the driver.