[HELP] NVEnc Settings

[tomek398]

I've been using NVEnc to convert my video files to H.265 format for archiving purposes. 

I try to understand various NVEnc setting available in Hybrid. Here are my questions:

1. Weighted Prediction
   

I have found following info:

Weighted prediction is not supported if the encode session is configured with B frames.

Weighted prediction is not supported if DirectX 12 device is used.

https://docs.nvidia.com/video-technologi...index.html

I don't understand what means that "session is configured with B frames" if the context of Hybrid settings (my setting are attached)
I've noticed that I can select  /P-frame Settings/Weighted P-frames even is some setting for B-frames are also active. Does it mean that, when I use Weighted Prediction then B-frames are not enabled?

I also don't understand second statement about Direct X 12

1. 2. I don't know if or when I should use B-frame ref mode: each/middle

Any help here would be welcomed.

[Selur]

The api can be accessed through different interfaces, since NVEncC does not use the DirectX interface can be ignored.
Regarding the weighted prediction of P-frames: I checked when B-frames are enabled, NVEncC will output:

weighted prediction with B frames unsupported.

So I'll adjust the gui to disable that option if b-frames are enabled. (I would recommend to use b-frames over weighted p-frames)

Cu Selur

[tomek398]

Thanks you for that explanation

Could you please also explain me that are recommendations regarding:  B-frame ref mode: each/middle

Which option: each or middle is recommended? I couldn't find good explanation on internet.

[Selur]

--bref-mode <string>
Specify B frame reference mode.

• auto (default)
  
• disabled
  
• each ... use each B frames as references
  
• middle ... only (Number of B-frame)/2 th B-frame will be used for reference
  

source: https://github.com/rigaya/NVEnc/blob/mas...ode-string
(atm. Hybrid does not have 'auto' as an option)

From the api, one can see what 'auto' would do:

B-Frames As Reference

Using B frame as a reference improves subjective and objective encoded quality with no performance impact. Hence the users enabling multiple B frames are strongly recommended to enable this feature.

To use the feature, follow these steps:

• Query availability of the feature using NvEncGetEncodeCaps API and checking for NV_ENC_CAPS_SUPPORT_BFRAME_REF_MODE in the return value.
  
• During encoder initialization, set NV_ENC_CONFIG_H264/NV_ENC_CONFIG_HEVC/NV_ENC_CONFIG_AV1::useBFramesAsRef = NV_ENC_BFRAME_REF_MODE_MIDDLE:
  • For H.264 and HEVC, this will set the (N/2)th B frame as reference, where N = number of B frames. In case N is odd, then (N-1)/2th frame will be picked up as reference.
    
  • For AV1, this will set every other B frame as an Altref2 reference but for the last B frame in the Altref interval.
    

source: https://docs.nvidia.com/video-technologi...-reference

So for H.264 and HEVC NVIDA by default would use 'middle' and for AV1 they would use 'each'.
Personally, I use 'each' for both, but that's not really based on any tests, but on sheer lazyness.

Cu Selur