Or triple, or more! All you need is a little parallelization, and an expensive graphics card, or multiple cards. Yeah, it’s not for everyone, but if you happen to have the hardware already, why not get the most out of it that you can?
What you need
You will need a video card that has multiple encoder chips on it by default, like a Nvidia 4070 ti or higher. For Nvidia you can look on this chart and see if it lists your card having more than one total number of NVENC chips.
You’ll also need the right software to encode it with, you can try it out now with the latest version of FastFlix!
With the latest FastFlix, it can automatically detect and download the required encoders to get this working. Make sure to select one of the rigaya encoders from the list, like NVEncC!
Split vs Parallel Encoding
First off, this is all possible thanks to rigaya and their awesome hardware encoding tools, and these are terms they are using in the tool itself. Parallel encoding works by splitting the incoming file into smaller chunks and working on them at the same time in separate threads.
In contrast, split encoding breaks up each frame into parts, so it’s better to use on higher resolution videos. Here is a handy visual they include in their code’s encoding settings to show the difference.
There are some caveats to using either. First, split encoding only works on Nvidia cards, whereas parallel works on Intel, AMD or Nvidia.
Parallel encoding of course has some restrictions as well as higher ram usage.
Video Quality
The documentation for parallelization does warn that split encoding may have a quality trade off, where fully parallel mode should be minimal loss.
I was surprised to find out in my testing that parallelization resulted in slightly higher quality videos!Though all are very very close to each other, within tenths of a VMAF percent each other. Overall I wouldn’t worry about quality loss or improvement.
The Butterfly video was run with high CQP (aka lower quality) and a few other different settings than the other videos. But each video had the exact same settings against themselves except for the type of parallelization.
Here’s a bit more zoomed in version of the chart:
Video Speed Improvements
As you can see from the chart at the top of the article, on an RTX 5090 which has three NVENC chips, is about twice as fast in split mode, and up to three times as fast in parallel mode.
Here’s a slightly different representation of the same data, mostly in case the charts don’t load.
In two of the cases the split and parallel encodes were very close on time, but parallel was always a bit faster. In the case of the Butterfly video parallel took a clear lead, down from 104 second encode to just 37 seconds.
Taking a peek in task manager, we can see the parallel encode fully saturates all the video encoding cores on the GPU.
Video Sources
- Butterfly 4K HDR https://codecalamity.com/projects/benchmark-videos/#butterfly
- Christmas Surprise 4K HDR https://codecalamity.com/projects/benchmark-videos/#christmas-surprise
- Amazon Educational Reel https://media.xiph.org/video/derf/
