Possible issues with cudaMallocManaged() when used wtih a vGPU?


we have code written for usage on a system supporting unified memory functionality (e.g. jetson agx). The used cudaMallocManaged() method ensures that the code will run on a system without unified memory, too. But what about cudaMallocManaged() when using vGPU (e.g. 2Q on RTX 6000)? Are there any issue that might occur on a virtualized GPU?

Many thanks in advance!


Currently, that page says:

The following CUDA features are not supported on vGPU:
Unified Memory

(This could change in the future)

So using cudaMallocManaged() with vGPU will throw an cudaErrorNotSupported, right?

Thank you for your fast reply!