[FFmpeg-user] ffmpeg nvenc without cuda

[Dennis Mungai]

On Fri, 20 Sep 2019 at 00:21, Johanna Nilson <jnils75 at gmail.com> wrote:

> Dear Sirs
>
> Is it possible to use nvenc (h264_nvenc and hevc_nvenc) without cuda at
> all? Just now I get error that No cuda device detected, when use something
> like:
>
> ffmpeg -f gdigrab -i desktop -framerate 30 -tune zerolatency -r 30 -c:v
> h264_nvenc -f mpegts udp://...
> ffmpeg -f gdigrab -i desktop -framerate 30 -tune zerolatency -r 30 -c:v
> hevc_nvenc -f mpegts udp://...
>
> But my system has architecture where there is nvenc support, but no cuda
> support at all. So I need to perform encoding at nvenc without any requests
> to cuda cores. Please, tell me, is it possible? Thank you very much



Hey there,

A few notes:

1. NVENC is a discrete silicon IP block on your GPU. It’s either there (if
supported on the GPU) or not. Cards prior to Kepler, including rebranded
mobile GPUs from the Fermi architecture do not have NVENC on them. The same
also applies to the Lower tier Gx1x8 GPU lines such as the GT 1030 and the
Quadro P500. If you have one of these cards mentioned, you cannot use NVENC.

2. Typically, when ffmpeg reports the “no NVENC device found” error, then
the command line in use is calling up encoder arguments not compatible with
NVENC on your current GPU. Common examples: Specifying an unusable pixel
format, or attempting to encode interlaced content on Turing.

3. If you’re specifically seeing “no CUDA device found” then confirm that
your device driver is indeed installed. The nvidia-smi tool can confirm
this.

For the reason in 2 and 3 above, please provide the full console output
from ffmpeg. To verify if (1) applies to you, also provide the output from
nvidia-smi.
I’m assuming that you have the device driver installed at the very least.

>
>