halcyon dirge
Hi, we're currently trying to get ALVR to run on our Desktop and use our NVIDIA Graphics card(RTX 2070 Super).
Unfortunately the stream immediately crashes upon the client connecting with the following error messages:
12:49:18.907 [INFO] Using NvEnc encoder
12:49:18.909 [ERROR] Encoder: cu->cuMemcpy2DAsync(&cpy, cuda_dev->stream) failed
12:49:18.909 [ERROR] Encoder: -> CUDA_ERROR_INVALID_VALUE: invalid argument
12:49:18.909 [ERROR] Encoder:
12:49:19.225 [ERROR] error in encoder thread: Failed to transfer Vulkan image to CUDA frame Generic error in an external library
The connection works when we switch to Software Encoding.
CUDA also works on our device for other applications(e.g. KoboldCPP, LMStudio) and the ffmpeg encoder example from the wiki using h264 works without problems(and uses the GPU).
What could cause this and how we could we debug this further? There's an open ticket on nixpkgs and the ALVR repo but nobody from the Nix crowd seems to have an idea how to debug this.
Could someone point is in a general direction of where in the tech stack the issue may lie? We've built it in Debug mode and are planning to add some more verbosity to the NvEnc C++ code to figure out the cause, but a rough direction would be really helpful. ALVR contains a lot of differernt tech layers that are still hard to wrap our head around.
The Nix buildfile looks like it mostly follows the Building from Source guide, although we're planning to double-check it's dependencies with the dependencies of other distros anyway.
The one discrepancy we could find is that they omit the cargo xtask prepare-deps --platform linux step. Unsure how important this step would be, as the build process runs without any errors.
dapper narwhal
umbral summit
last bough
