When attempting to run GPT4All with the vulkan backend on a system where the GPU you're using is also being used by the desktop - this is confirmed on Windows with an integrated GPU - this can result in the desktop GUI freezing and the gpt4all instance not running.

I believe the problem is contention in GPU memory between what the desktop is trying to use and what we are trying to allocate with vulkan.

I can reproduce this on my old Windows 10 laptop with the integrated AMD Vega GPU consistently

Not sure how to fix at this time :(

Reproduction

Reproduced with old Windows 10 laptop trying to use the integrated AMD Vega gpu. More system details to come.

Expected behavior

We should detect this case ahead of time and fallback to CPU.