Update to 4K RedCar HEVC benchmark

Howard-Vigorita wrote on 12/12/2025, 10:04 AM

I did an update to the media substitution used in the RedCar 4K HEVC benchmark which pretty closely models my multicam HEVC workflow. It demonstrates how dramatic the performance penalty can be splitting and overlapping media clips in Vegas and/or using 4:2:2 clips instead of 4:2:0.

On my 4090 and 5080 systems, it yields better performance with 3 independent 60mbps 4k 4:2:0 HEVC tracks than the original which used 3 independent HD tracks: m2ts@13 mbps, mp4 @ 24 mbps, plus a 50mbps HD422 XDCAM. The zip also includes my old version with a media substitution scheme that referenced the same HEVC clip 3 times instead of using 3 separate copies... which is like splitting a clip in Vegas and overlapping it on itself. Which turns out to run quite a bit slower, especially on vp23 b302. The updated zip file can be downloaded here:

https://drive.google.com/file/d/1oNyunKSjcdUDHqKLvL-7V6ohDV1Jj24K/view?usp=drive_link

In order to play or render the new version, the included BAT file needs to be run to make 2 additional copies of the supplied HEVC clip. The included VEG projects are identical to the Sony original other than being saved after media substitution with vp16 so they'll work on all subsequent versions of Vegas.

Here are the render-time results I got on my 2 desktop workhorse systems:

Comments

Howard-Vigorita wrote on 12/12/2025, 10:20 AM

Also note that Vegas isn't yet able to decode 4:2:2 hevc with an Nvidia 5080 although it can do it with an Intel Arc. I don't have an Intel Arc in a machine at the moment so I'm not able to test that right now. If someone has an Arc running, transcoding the supplied 4:2:0 hevc clips to 4:2:2 would be interesting.

johnny-s wrote on 12/14/2025, 8:40 AM

@Howard-Vigorita

I'm a firm believer in the less is more philosophy.

If this suits ...

Can u upload a single project file with all the converted and to use source files only, not for 3 projects, tests.

An included batch file to be run first to create any necessary source files is ok.

Can u include jpg of render template and project properties, latter may be redundant as project file may automatically set as required.

I have currently my Intel a770 removed from PC.

I can then give render times for before and after reinstalling the Intel gpu for both vp22 and vp23.

Howard-Vigorita wrote on 12/14/2025, 10:59 AM

You can use the link provided to download the zip and extract it to an empty folder. It has 2 project versions, all the ancillary files, and one hevc-mp4 which the bat file copies to make 2 more. I don't have enough cloud storage to put 3 copies of the mp4 in the zip. Btw, you don't need to run the bat file to try the OLD version... it only uses the one 60mbps 4k 4:2:0 hevc in the zip but runs much slower than using 3. The original Sony HD version used a 50 mbps HD422, but the other 2 were low bit-rate 4:2:0 transcodes.

The Sony 29.97 fps HD project config matches the media frame-rate and is embedded in the projects... I didn't mess with them other than to substitute 4k Hevc media at the same frame rate. You can inspect the project settings in Vegas after you load it. The render times I posted are for the internal-default Magix HD 29.97 Nvenc Avc preset... which looks like this:

... but you can compare results with others like MainConcept, Qsv, or Vce. Or other plugins like Voukoder. Gpu renders of hevc 4:2:0 media will be faster than MainConcept but they all seem to be substantially slower with 1 clip compared to 3.

Howard-Vigorita wrote on 12/30/2025, 10:24 AM

Last night I tried an hevc422 media transcode with the RedCar benchmark... and locked up vp23 b356 with Vegas set to gpu = Nvidia 5080 and decoder = auto (5080).

Also tried it on an Ultra9 275hx Nuc with 5080 and got same result with same settings. The onboard Intel igpu is able to decode hevc422, however, but not by just selecting it in i/o. I could get it to play and render correctly with gpu=5080 & i/o = off. Also with i/o = exp-hevc & Intel igpu. Or with gpu = Intel & i/o = auto (Intel). And observed significant decoding utilization on the igpu when selected. But render times for the workarounds were 3x to 6x longer than hevc420.

Btw, I couldn't figure out how to do Nvidia or Intel transcodes to 10bit hevc422 with ffmpeg. Closest I could get was 8bit 422 w/qsv. Nvidia would only do 420 or 444. Had to use libx265 via cpu. Don't know if this is due to gpu limitations.