Runtime: cudaErrorInsufficientDriver from CUDA 9.1 Sample Code

I currently have a P40, with 385.41 driver (Dell R740xd, VMWare ESXi 6.5).

The sample CUDA 9.1 code compiles fine, but I get cudaErrorInsufficientDriver at runtime from cudaGetDeviceCount.

When I try to install the 385.54 or 390.85 drivers, it tells me that I don’t have any compatible hardware.

Is the P40 supported for CUDA development?



Yes, the P40 is supported for CUDA development.

If you are using the P40 as a shared graphics device for VMWare VMs, it may or may not be usable for CUDA depending on a number of configuration and use parameters.

It would be important to know what you are doing with VMWare virtualization (if anything).

If you are using GRID software for vGPU virtualization/sharing, you should probably ask your question on the GRID forums

If you are doing GPU passthrough, you should probably indicate that.

Thank you - the P40 is set to “Shared Direct”.

Is there a document that lists the config/use parameters? I’d love to try to get a few samples running and tested before GTC next week…

I got some support at GTC. For some reason my VDI will not support a driver above the 386.09 that I have installed now. Version 391.29 tells me that it can’t find compatible hardware.

Anything below version 387 will only support CUDA 9.0. CUDA 9.1 requires 387 and above.