V3 Still Slower Than V1

  • First of all, thank you Vouk for your awesome encoder and for making your work available to the community for free. My buddy and I have been using it for over a year now, and we're eternally grateful to you.

    With regards to the subject of this post, the aforementioned buddy and I have been keeping an eye on development progress of V2 and now most recently V3, and unfortunately we are still stuck on V1.2.1 due to the significant performance decrease of later revisions. We've been leveraging a couple of standardized project files between the two of us for the sole purpose benchmarking our performance as we make various hardware changes to our PC builds. His system is based on a Ryzen 3900X 12-core CPU with an nVidia 1080 Ti GPU, while my system is based a ThreadRipper 3970X 32-core CPU with an nVidia Titan X (Pascal) GPU.

    Benchmark #1 with Voukoder 1.2.1

    - Source File: C:\Users\root\AppData\Local\Temp\Bench_14_3.prproj

    - Output File: C:\Users\root\Desktop\PeePee Test_1.mp4

    - Preset Used: KT Editing

    - Video: NVIDIA NVENC hevc encoder, 1920x1080 (1.00), 23.98 fps, 00:02:03:10

    - Audio: AAC (Advanced Audio Coding), 48000 Hz, Stereo

    - Encoding Time: 00:01:11

    Benchmark #1 with Voukoder 3

    - Source File: C:\Users\user\AppData\Local\Temp\Bench_14.prproj

    - Output File: C:\Users\root\Desktop\PeePee Test.mp4

    - Preset Used: KT Editing - Vouko V3

    - Video: 1920x1080 (1.00), 23.98 fps, 00:02:03:10

    - Audio: 48000 Hz, Stereo

    - Encoding Time: 00:01:42

    Benchmark #1 with VRPT 1.3


    Benchmark #2 with Voukoder 1.2.1

    - Source File: C:\Users\user\AppData\Local\Temp\Bench-2.0.1_14_4.prproj

    - Output File: C:\Users\user\Desktop\PeePee Test 2.0_1.mp4

    - Preset Used: KT Editing

    - Video: NVIDIA NVENC hevc encoder, 1920x1080 (1.00), 23.98 fps, 00:02:37:01

    - Audio: AAC (Advanced Audio Coding), 48000 Hz, Stereo

    - Encoding Time: 00:00:49

    Benchmark #2 with Voukoder 3

    - Source File: C:\Users\user\AppData\Local\Temp\Bench-2.0.1_14.prproj

    - Output File: C:\Users\user\Desktop\PeePee Test 2.0.mp4

    - Preset Used: KT Editing - Vouko V3

    - Video: 1920x1080 (1.00), 23.98 fps, 00:02:37:01

    - Audio: 48000 Hz, Stereo

    - Encoding Time: 00:01:22


    Benchmark #2 with VRPT 1.3



    Please don't hesitate to let us know what other tests you'd like us to run or log files you'd like us to send. We're both very happy to help debug the performance issue with you.

  • Same here,

    It seems the R3 is slower for me than the R3rc2. When I say slow I mean it literally takes 8 hours for encoding a x264 8 min long video (on an i7 7700k) where it tooks around 25 min before.

    However this problem only occurs in Premiere Pro and not in Media Encoder, for me when I launch Media Encoder and add a video manually then launch the encode, it is working perfectly.

    • Offizieller Beitrag

    Wickyed Thats really strange. I just went through the changes between both versions and not many things have changes there. At least nothing regarding rendering and encoding.

    PP_User Alright. I'll take a look at it again. But is it possible you compared yuv420 8 bit (nv12) in VRPT with yuv420 10 bit (p010)? If voukoder is working in non 8 bit mode it does a 2 stage framebuffer conversion that is not very optimized yet. What are the results of v3 if you use yuv420p instead? Please send a logfile of it.

  • Thank you for the quick reply, Vouk!

    To be completely honest, I am brand new to VRPT and left all settings at default. I see that there are dozens of options under Pixel Format, but wasn't sure which one you wanted me to select. It is currently defaulted to "Any". That being said, I only included the VRPT results because I noticed you asked for it in another thread. I was more concerned about the results of the two benchmark projects showing consistent performance degradation in versions newer than 1.2.1.

    That being said, it would appear you have identified a primary factor in the issue. Our benchmark presets are all configured to use 10-bit depth.

    Here are the updated benchmark results when downgrading depth to 8-bit:

    8-bit Bench #1 | Voukoder V1 = 1:11 | Voukoder V3 = 1:22

    8-bit Bench #2 | Voukoder V1 = 0:50 | Voukoder V3 = 0:46

    Interestingly, the results are mixed.

    With benchmark #1, performance of Voukoder V3 is improved from 1:42 to 1:22 when downgrading depth from 10-bit to 8-bit, but is still slower than Voukoder V1.2.1 at 1:11.

    On the other hand, with benchmark #2, performance of Voukoder V3 exceeds V1 when downgrading depth from 10-bit to 8-bit, improving from 1:22 to 0:46, which is faster than Voukoder V1.2.1 at 0:50.