EposVox: NVENC vs AMF/VCE vs QuickSync vs X264 - ULTIMATE Encoder Quality Analysis 2020

Yeah surprised me as well, I remember when using gfx cards encoding was fast, but quality sucked.

This has a pretty big implication. As long as you can get tools like Handbrake to use NVenc, then the CPU is kind of irrelevant for Video encoding. Video encoding is one of very few common Highly Parallel use case for multicore CPUs. For most people it's the only one.
 
I use the NVENC encoder on a gtx 750ti, paired with a i5 4590s, in order to run OBS to stream to facebook live. Since it is a live camera stream with slides (church service), I was very surprised at the efficiency and quality of the stream.

In the previous weeks I was using my own personal SFF rig with a Ryzen 2600, and Fury Nano, trying both the x264 encoder and the AMD AMF encoder, and the less powerful PC is much better and more efficient at managing the stream.

The first party support by Nvidia to work on the NVENC encoder in OBS really shines through. It is a shame that AMD does not have the spare personnel to dedicate to programs like this.
 
They all suck. Use CPU h265 its significantly better. Do it when youre goong to sleep and it will be done when you wake up.
At this time the main use case for h.264 based encoders should be real time encoding (e.g. streaming) and backwards compatibility situations. h.265 or HEVC, as tangoseal points out is much better.
 
The video posted by the OP presented benchmark results show that Turing's NVEC does really well in HEVC real-time encoding for streaming in terms of compression vs. quality ratio.
 
Good call I was skimming the early sections and skipped to the end and missed the mixed comparisons of x264 vs NVENC HEVC /NVENC h.265 (and navi encoder) - a non-intuitive comparison. I would think you'd test x265 vs NVENC HEVC for an apples to apples, although that is a long video maybe he explains it somewhere. I personally find these 25+ min videos with tons of graphs and limited time indexes a misfit for the youtube video format. Just post an article so people can skim and focus on the graphs they care about and read the analysis in depth.
Overall interesting results though, the Nvidia hardware encoders are doing well now even with NVENC h.264 compared to x264 slow preset, which is a good quality h.264 encoder but as the name implies.. slow. HEVC has a ~15-50% efficiency benefit over h.264 so you do expect it should do better when compared that way.
 
Good call I was skimming the early sections and skipped to the end and missed the mixed comparisons of x264 vs NVENC HEVC /NVENC h.265 (and navi encoder) - a non-intuitive comparison. I would think you'd test x265 vs NVENC HEVC for an apples to apples, although that is a long video maybe he explains it somewhere. I personally find these 25+ min videos with tons of graphs and limited time indexes a misfit for the youtube video format. Just post an article so people can skim and focus on the graphs they care about and read the analysis in depth.
Overall interesting results though, the Nvidia hardware encoders are doing well now even with NVENC h.264 compared to x264 slow preset, which is a good quality h.264 encoder but as the name implies.. slow. HEVC has a ~15-50% efficiency benefit over h.264 so you do expect it should do better when compared that way.

HEVC isn't the point, since it's all about streaming, and streaming services don't support it.

Maybe someone else looks more into the offline potential.
 
Well, the TL;DW of it is that Turing's NVENC on many of the tests, still does quite well particularly with the popular streaming x264 encoding format in terms of objective and subjective quality at various bitrates and game genres even compared to Navi's VCE x265 encodes. In the conclusion, it was pointed out that Nvidia and Intel has better software support for OBS compared to AMD which is virtually non-existent at the moment and he hopes that future iterations of NVENC, Quicksync and VCE will also support AV1 and VP9.

The author also provided a link in the description for the written article: https://eposvox.com/post/ultimate-encoder-quality-analysis-2020-nvenc-vs-amf-vs-quicksync-vs-x264

Interestingly, Twitch broadcasting guidelines have specific settings for NVENC but not QuickSync nor VCE. https://stream.twitch.tv/encoding/

Encoding Quality
x264 offers a wide range of presets that change quality significantly, and presets above Faster require CPUs with 6+ cores. NVIDIA NVENC offers consistent quality based on the generation of the encoder. The updated NVIDIA Encoder (NVENC) on Turing-based NVIDIA GeForce GPUs (RTX 20-Series and GTX 1660/Ti) will typically produce superior quality than x264 Fast and on par with x264 medium. While the older generation (Pascal, Kepler) are similar with veryfast/faster quality.
 
Last edited:
This is kinda pointless to compare HEVC encoding, as none of the live-stream services accept it.

Why not compare VP9 encoding, instead?
 
AV1 would be a good addition it’s already well supported and growing fast, open source/freely licensed, and has a similar bump in efficiency over h.264 and VP9.
 
Back
Top