Why b frames not does less file size ?

[Aug]

Thanks for reading my topic.
Here is Linux Lubuntu 24.04.3 kernel 6.14 using a GT 640 driver 470.
I see in some sources b frames does better compression than I and P frames ... well .. in that logic the bitrate is less than P frames, but not are being exactly it.
When converting using nvenc the file is the same size with and without b frames even using a good value for GOP.
In Nvidia Nvenc doc has the information
"B-frames are great because they increase image quality, but they consume a lot of your available bitrate, so they reduce quality on high motion content"
If b frames does better compression had to be a less file size.
I see another issue using b frames ... in scenes with much motion and fade in out does issues in video quality.
Really using b frames for video archiving is a good solution or is better only using I and P frames ?
I see much artifacts when converting from xvid to H264 too.
Have a nice week.

[jagabo]

If you are using bitrate based encoding the file size will always be the same because

Stream size = Bitrate * running time

If you reduce the number of b-frames you will have more p-frames. And less bitrate will be used on each p-frame, reducing the overall quality.

[pandy]

b (B) frames use information stored in frames that precede current b frame and information from next frames (I,P,B) - as such B frames depends on other frames - as such they offer highest compression - as B frames can offer higher compression then higher bitrate can be used for I and P frames (thus increase overall quality).

[dannyboy48888]

B frames is a rabbit hole since the early ages of video encoding. Depending how they are used they can improve quality or degrade it. For archiving I usually dont use them as the savings is marginal for me in mpeg-2 compared the effort for good quality that's needed. But for viewing they can increase the perceived quality as you can have a slightly lower frame that is smaller but go unnoticed as other frames don't reference it. In avc+ codecs they can be used as references. When too many (3 or more) are used you may see the fading issues etc you described. Try no b-frames then 1. And dont set them to the same q as the other frames as you'll get pulsing as they end up being better quality than the I or p frames. they have 2 frames they reference so with that they are designed to have a slightly higher q for the same quality. Simplification I know and I'm sure some will chime in to correct this but just the primer is all. In ye olden times they also required more memory to the point that not using them was the only way to keep sync etc, but that's not the case now and days.

[pandy]

B frames is a rabbit hole since the early ages of video encoding. Depending how they are used they can improve quality or degrade it. For archiving I usually dont use them as the savings is marginal for me in mpeg-2 compared the effort for good quality that's needed. But for viewing they can increase the perceived quality as you can have a slightly lower frame that is smaller but go unnoticed as other frames don't reference it. In avc+ codecs they can be used as references. When too many (3 or more) are used you may see the fading issues etc you described. Try no b-frames then 1. And dont set them to the same q as the other frames as you'll get pulsing as they end up being better quality than the I or p frames. they have 2 frames they reference so with that they are designed to have a slightly higher q for the same quality. Simplification I know and I'm sure some will chime in to correct this but just the primer is all. In ye olden times they also required more memory to the point that not using them was the only way to keep sync etc, but that's not the case now and days.

Firstly you can have adaptive frame type (so your encoder is aware of difficult content and insert I or P frame instead B frame) and secondly that's why you need good RDO ( https://en.wikipedia.org/wiki/Rate%E2%80%93distortion_optimization?useskin=vector ) - usually decent encoders provide both techniques.

[Selur]

Afaik. the (12years+ old) GT 640 does not use nvenc, but the old nvcuenc cuda encoder which only supported 1 fixed b-frame, not multiple and adaptive b-frame placements (https://github.com/Selur/CudaEncoder has the sdk doc in the docs folder)
=> looking at the nvenc doc seems wrong

Cu Selur