Loking to improve NVENC H265 render speed

marchwill wrote on 2/25/2023, 3:38 PM

Hey folks! Long time user of Vegas here (since version 7), first time poster. I'm looking to share some experiences and learn from you, hoping to increase my render speeds.

I have several PC configurations in my studio, and the newest one is a 13700k with a RTX 4080. I did some render tests on it with Davinci Resolve Studio and came to an amazing 163 FPS while render 4K footage in a multi cam project with color correction. That's 11 seconds per minute of video, which is amazing.

After this, I decided to come ask you guys how's the experience been with Vegas renders. On this same setup, I can manage to reach about 50fps. But on a more realistic scenario, with adjustments and color correction, that number stays below 30 fps. This is on the setup I listed below. On a 12900K with RTX 3080, the number is about 10fps lower.

This is using Vegas' native render options. The main thing I'd like to address is the NVEC stuttering. While rendering, the meter keeps going as high as 150fps, which would be on par with Davinci results, but it drops really low every few seconds. This results in a very low average fps. Is there any way this could be fixed in a future update? I've read about it somewhere else, are the devs aware of his particular issue? If this gets fixed, I'd have a huge boost in performance.

For comparison, Voukoder (while it was still working) doesn't show the average fps while rendering, but I did the math counting frames vs. time and it's about 50% faster. It does provide a smooth render speed, without stutterings, but doesn't reach a very high fps performance.

So what are your thoughts on this? I'm thinking about building a benchmark project if anyone wants to test on their own so we can get a better look at this situation.

 

 

Comments

Howard-Vigorita wrote on 2/25/2023, 4:11 PM

Both Voukoder and DaVinci use lower res math and storage so to some extent they trade quality for performance. At least that's what the objective transcoding analysis I've done with ffmetrics suggests. Oddly, ffmpeg itself does not have this problem when used directly which I do allot. However the ffmpeg libs they employ, make them highly optimized for Nvida gpus. I run both Vegas and Resolve Studio but I'm not proficient enough with Resolve to do multicam. Much better editing my mostly 4k hevc concert shoots with Vegas. If you are good at it with Resolve, you probably want to stick with that. I find my best performance/quality/cost mix is with high-end Amd gpus with Intel Arc doing the decoding in Vegas. Fwiw, Resolve can do Intel Hyper rendering with multiple Arcs which Vegas cannot do yet. I've never managed to score a high-end Nvidia gpu so I don't have a personal direct performance experience with that. I am interested in the new 4000 series so I might give it another go at some point.

If you're interested in a Benchmark project, check out the thread that is pinned above which leads to this. I did a test conversion of that project to vp19 and it was a learning experience for me. Don't know if it would be possible to convert something that sophisticated to Resolve. But if you have a benchmark already written in Resolve, it might be interesting to convert it to Vegas.

Former user wrote on 2/25/2023, 10:27 PM

I have several PC configurations in my studio, and the newest one is a 13700k with a RTX 4080. I did some render tests on it with Davinci Resolve Studio and came to an amazing 163 FPS while render 4K footage in a multi cam project with color correction. That's 11 seconds per minute of video, which is amazing.

For comparison, Voukoder (while it was still working) doesn't show the average fps while rendering, but I did the math counting frames vs. time and it's about 50% faster. It does provide a smooth render speed, without stutterings, but doesn't reach a very high fps performance.

For encoding, I think the main problem is the render engine, it doesn't have the efficiency of modern NLE's. The hardware encoder can't be faster than the rendering engine, in case of Voukoder I think it does best it can, if Vegas could render frames faster it could encode frames faster.

This shows the efficiency problem. I turned off the decoder in both Vegas and Resolve. The source file is 4K AVC, the rendered file 720P NVENC. In resolve I could keep project as 4K, but for Vegas via Voukoder I had to change it to 720P, I don't think this is cheating on behalf of Vegas because the 4K file still has to be decoded the same, and also if you encode via internal vegas settings to 720P even if you have a 4K project the project settings change to 720 before the encode.

Anyway these are the results,

Vegas 115fps at 100% CPU (12 core 5900x)

Resolve 190fps CPU 75%

 

My example is more of an apples to apples. but yours is not quite. It's because of the dual HEVC encoder on 40 series GPU's, It works with Resolve but apparently doesn't work on Vegas. Keep in mind the bucket analogy, this dual encoder helps resolve get more water to the bucket faster but maybe not for Vegas because pipeline to the bucket can not supply enough water.

It's still good that Vegas did use 100% of my CPU, it's the obvious bottleneck on my system with this test, but efficiency not so good.