Voukoder 6 Export is very slow

  • I've recently got the Voukoder 6, and I got it working on Premier Pro. My Drivers are up to date (I read all the install instructions, and everything). Now I'm trying to export a Video that is about 1 Hour in length in 4k HEVC NVENC (Lossless Settings). I've got 2 RTX2080s in SLI, but Windows reports less than 20% utilization on both the cards when I'm using Voukoder 6 and the export time is about 4.5 Hours for the 1 Hour video. But when I use Voukoder 1.2.1 for the same exact settings, the export time is a little less than 1 Hour, over a 4x increase and Windows reports a 99% usage on GPU_0 and about 30% usage on GPU_1.

    The only difference in settings is which Version of Voukoder I use. Everything else I've matched across the board.

    One thing that I did notice in V6: When I tried to select my second GPU (GPU_1) to use as the Rendering GPU (I was doing it just to see if there'd be any difference in performance), it seems to ignore which GPU I select as the workhorse. Although I think it's probably because of my SLI settings.

    I'm not sure if this is a bug or not, but I'll be sticking to V1.2.1 for now. I love this Transcoder, but I'm not sure the newer versions are able to compete with the raw renderer in Adobe Premier Pro. Render times is about the same between Voukoder and Adobe for h.265 HEVC rendering.

    I did let it render for about 30 minutes thinking that maybe it was miss predicting the timings. But, no, it was rendering at a comparable rate to Adobe Software Rendering.

    My System Specs if it helps:
    RTX2080 x2 SLI

    16GB DDR4 @3Ghz

    Win10 Pro (1909, 18363.1016)

    Nvidia Gaming Drivers 452.06

  • Vouk

    Added the Label In Progress
  • The export is slow, because Voukoder uses FFMPEG, which does not allow internal trascoding (within the video card) process. Rigaya's NVENC can be much faster because of internal encoding. Howeve internal encoding disclude the possibility of video manipulation with video editor softwares.

  • Hello.

    The same situation. Doesn`t matter the size of video. Coding takes only 20% of Nvidia recourses. I try all releases of Voukoder and found that Voukoder, working via Connector (from ver 2.3) gives this result, but with 2.2.1 and later if we work as premiere plugin, without connector works perfect (90-100% of NV encode) . May be there is some solution to make it works with Connector?

    Premiere pro NVidia latest Studio 452.06

    Xeon 2678v3 3.3 unlocked, X99 + C610 Raid, 32Gb DDR3 ECCreg, RTX 2060 Super 8Gb, 500Gb NVMe + 1Tb (Raid0 SSD 2x500)

    Edited 2 times, last by Allone ().

    • Official Post

    If Voukoder Performance Analysis is TL, DR here:

    When looking at your Voukoder logfile we'll most likely see high "vRender" times. This is the time Premiere needs to render a single frame. As the whole export is a synchronous chain the final encoding fps can be any higher than a single part of the chain.

    You might want to add your log, but I guess it'd most likely confirm this.

    Even more DR:

    Your GPU could be able to encode with like 1000 fps, but if premiere is only able to deliver 10 fps the usage would only be like 1%.

    P.S.: I don't know why an older version might be faster. This should have been improved with the connector 1.5.0.

    • Official Post

    You are right. This is a rare case (and a good example) where actually the video encoding is the bottleneck.

    [17:12:11] Frame #233: vRender: 20 us, vProcess: 10567 us, vEncoding: 29849 us, aRender: 63 us, aEncoding: 793 us, Latency: 42414 us
    [17:12:11] Frame #234: vRender: 20 us, vProcess: 10604 us, vEncoding: 29452 us, aRender: 63 us, aEncoding: 442 us, Latency: 41711 us
    [17:12:11] Frame #235: vRender: 19 us, vProcess: 10465 us, vEncoding: 29521 us, aRender: 64 us, aEncoding: 462 us, Latency: 41812 us
    [17:12:11] Frame #236: vRender: 20 us, vProcess: 10410 us, vEncoding: 29694 us, aRender: 61 us, aEncoding: 10 us, Latency: 41313 us
    [17:12:11] Frame #237: vRender: 19 us, vProcess: 10273 us, vEncoding: 29524 us, aRender: 62 us, aEncoding: 446 us, Latency: 41454 us

    As I can see you are encoding in 10 bit. That means:

    Step (per frame) Time (in ms) Theor. FPS
    vRender - Premiere renders in "VUYA 4:4:4:4 (Float)" 0.02 50.000
    vProcess - Pixel format conversion to "p010le" + 10.50
    vEncoding - Actual GPU encoding + 29.50 24
    Total 41.50

    You should get a huge performance boost if you render in 8 bit. Do you have to use 10 bit?

    As far as I know the old version of Voukoder did not do a proper 10 bit encode. That's most likely why they were faster.

  • Vouk

    Set the Label from In Progress to Not a bug