Question: Does this NVENC patch work with VGPU Unlock?

[gogito]

Hello,

My question is:
Does this NVENC patch work with VGPU Unlock?

2060 Super 8GB (PVE host) (Grid Driver 550.163.02), trying to install nvenc patch failed because this driver is not in supported list.

-> Debian12 VM (Recognized as GRID RTX6000) (Guest Grid Driver 550.163.01), also installed the nvenc patch.

I gave it 4GB using toml file. I ran the nvidia_patch for more stream on the guest VM. For the host, it says doesn't support my driver version which is 550.163.02. Only 550.163.01 is supported.

I can't start more than 7 process of ffmpeg. I am running test directly on the VM, no docker or any other layer.

This is the command I use to test, each takes around 100MB of VRAM
ffmpeg -y -vsync 0 -hwaccel cuda -hwaccel_output_format cuda -f lavfi -i testsrc -t 5000 -c:a copy -c:v h264_nvenc -b:v 1M -f null -

When I try to start an 8th ffmpeg process, I get this

[AVHWDeviceContext @ 0x55b0c3641180] cu->cuCtxCreate(&hwctx->cuda_ctx, desired_flags, hwctx->internal->cuda_device) failed -> CUDA_ERROR_OUT_OF_MEMORY: out of memory
Device creation failed: -542398533.
[rawvideo @ 0x55b0c3622240] No device available for decoder: device type cuda needed for codec rawvideo.
Stream mapping:
Stream #0:0 -> #0:0 (rawvideo (native) -> h264 (h264_nvenc))
Device setup failed for decoder on input stream #0:0 : Generic error in an external library

I already tried using 1 big ffmpeg command and it can consume the full 4GB just fine so it's not actually out of memory.

If I use the test script in patch tester, it works fine so I'm not entirely sure where the problem is or if the tester script is still accurate.

On the guest:

On the host