NVIDIA CUDA for non-native ffmpeg acceleration?

Help
1

I have successfully configured the latest proprietary NVIDIA drivers on a 3rd Gen GeForce Titan X.

The GPU supports HEVC encoding but no AV1, which I prefer.

I can encode video at 1x speed with a highly throttled CPU (as the CPU sucks and overheats otherwise, even with graphene thermal pad and a fine CPU fan).

But I would prefer encoding on the GPU.

Even though it doesnt natively support AV1, I suppose I could get that running with CUDA? I expect worlds better performance, would be cool to use it for more encodings.

I couldnt find many cuda packages for base support so I assume that is included in the driver. But what do I need to do for ffmpeg to support cuda?

Make a derivation and build it with CUDA support? I couldnt find anything here in the forum and after a quick web search so I am opening this thread.

Goal: run ffmpeg, maybe opusenc and other tools with CUDA on the GPU.

2

AFAIK there exists no such codec in ffmpeg that could encode to av1 using only the CUDA capabilities of your gpu.

2 Likes
3

Yeah there is no CUDA implementation of AV1. Either your GPU can do AV1 in hardware or you have to do it on the CPU (which i wouldn’t recommend, way too slow). If I may recommend a solution, get a first gen Intel Arc card. They’re cheap and good for AV1. Get the lowest tier model, the encoding speed doesn’t change between them, only the amount of simultaneus encodes which is capped by the amount of VRAM.