using cuvid in hwaccell ffmpeg

[Mimo]

hi,

i am trying to set the ffmpeg hw accel to use cuvid instead of dxva2.
can you add a configuration for it that i can select cuvid for nvidia card ?

[Selur]

will look into it

[Mimo]

will look into it

thx

[Selur]

I did send you a link to a dev version which allows to set 'cuvid' in ffmpeg hwaccel settings.
-> Did you test it?

[Mimo]

I did send you a link to a dev version which allows to set 'cuvid' in ffmpeg hwaccel settings.
-> Did you test it?

yes, working fine, but my GPU is not able to decode HEVC even with cuvid (860M feature set E - hybrid decoding) :

FFmpeg : cuvid hwaccel requested for input stream #0:0, but cannot be initialized.

but thanks ...

---

looks like cuvid is broken in my drivers as even AVC is stopping with same error ...

[Selur]

Ah okay.

Cu Selur

[HW decoder]

Ah okay.

Cu Selur

I have feature set H. Maybe I can test it better.

[Selur]

.. I already send you a link to the dev version via PM before you posted this, you just didn't check your PMs,...

[HW decoder]

.. I already send you a link to the dev version via PM before you posted this, you just didn't check your PMs,...

Hi!

I tested the new version.

During the installation an nvenc file is missing, which is not problem if you did not uninstall the previous version.

The Cuvid decoder is faster than the DXVA2 in Hybrid (80 fps vs. 40-50 fps in 1080p@24 Blu-ray and 1080p@24 HEVC materials) but it was not so fast like the "NVENC Cuda" and "NVENC native" decoders (over 500 FPS ) in StaxRip.  Staxrip had a GPU usage around 98%-100%  for the NVENC (AVC and HEVC encoding) + GPU decoding while Hybrid used the "VPE engine" only between 15%-25% for NVENC HEVC and AVC decoding and encoding. I monitored it in the "sensor panel of the  "GPU-Z" GPU information software, where you can see the usage of all important components of your VGA card. So Hybrid did not really use the card efficiently, while StaxRip used it to the max (99)

After the PC tests, I will make my "Laptop tests" too.

I think you can implement these faster decoders too in the Hybrid, because they work well with Avisynth V.synth and FFMPEG too.

Maybe, if you examine StaxRip, you can see and figure out how did they implemented them to their software. Can't you?

[Selur]

Maybe, if you examine StaxRip, you can see and figure out how did they implemented them to their software. Can't you?

Had a quick look, don't see how StaxRip should be faster when avs or vpy is used.
In case neither Avisynth nor Vapoursynth is used StaxRip seems to use (https://github.com/stax76/staxrip/blob/m...g/NVEnc.vb):

--avhw [<string>] use libavformat + cuvid for input
this enables full hw transcode and resize.
avhw mode could be set as a option
- native (default)
- cuda

and feeds NVEnc directly with the source, so no surprise.
No clue whether I add support for this, since I don't have the time atm. to try to figure out which inputs NVEncC supports directly.
-> might come but probably not any time soon, since I still have a long list of bug to fix and bugs have priority over new features.

Cu Selur