Posts by MyPOV

    There are two configurations, GTX1650 using old Volta encoding of cards of previous generation GTX10x0 which only offers an incomplete H265 without B-Frames (it is this essential point that we have to do real efficient H265) and those using Turing architecture with a full H265 with B-Frame; It is naturally these latest models that must be chosen since they are the gateway to H265 encoding. Note that performance is almost identical for all models using Turing since they have the same hardware for encoding, the slight differences arise from differences in memory access and their frequencies.


    GTX 1650 (first) : GPU TU117 Volta 4Go GDRR5 896 Cuda without B-Frame

    GTX 1650 Super : GPU TU116 Turing 4Go GDDR6 1280 Cuda with B-Frame

    GTX 1650 G6 : GPU TU117 Volta 4Go GDDR6 896 Cuda without B-Frame

    GTX 1650 : GPU TU106 Turing 4Go GDDR6 896 Cuda with B-Frame

    GTX 1650 : GPU TU116 Turing 4Go GDDR6 896 Cuda with B-Frame

    GTX 1650 Ultra : GPU TU106 Turing 6 Go GDDR6 896 Cuda with B-Frame

    All this is to be confirmed and you have to be careful when buying, especially because it is not necessarily well specified by the manufacturers of video cards.


    There may also be a difference in performance depending on the number of CUDA cores you need. But there is no method to evaluate it and it depends on the complexity of montage and number of effect used.


    source : https://forum.magazinevideo.com/topic/38776-infos-les-cartes-graphiques-nvidia/?do=findComment&comment=186484

    ??


    There are millions of people who want to upload videos created with Premiere to Youtube but have problems either because their file is too big or because final quality on Youtube is not enough good.

    What do you not understand in this simple concept of wanting to upload a video of a reasonable size to have a satisfactory rendering on Youtube?


    To say that there is no solution is not an acceptable answer; So I offer mine to the community,

    Hello, I disagree with your message...


    I'm talking about the weight / quality ratio for Youtube destination, so by also considering upload time which is a problem for many people, therefore lossless formats are irrelevant.


    No, the GTX1000 series is not the most efficient for the quality / weight ratio since it does not support B-Frames for HEVC. The object of the discussion is not the encoding speed which I consider sufficient with Nvenc.


    GTX 16x0 and RTX cards have almost same performance and quality for encoding,

    (and

    GTX 1650 Super with GPU TU116 4Go GDDR6

    GTX1650 with GPU TU106 4Go GDDR6

    GTX1650 with GPU TU116 4Go GDDR6

    GTX 1650 Ultra with GPU TU106 6 Go GDDR6

    But not

    GTX1650 (original) with GPU TU117 4Go GDRR5

    GTX 1650 G6 with GPU TU117 4Go GDDR6

    )

    for the smallest file size possible, use hevc yuv4:4:4, its in voukoder but for now you have to manually input a hundred settings in this post


    I do not see the point of doing 4:4:4 starting from 4:2:0 which is largely what is more popular on Youtube. The normal user is not going to do this, he will be content with the complexity of the settings of software like Handbrake.


    I think that you are lost in your settings and that you forget the main thing which is not to have fun with tons of parameters but to have an effective generic solution for the majority of users who want to quickly put a video on Youtube and get a good quality.

    After many tests and analyse about Handbrake's files, here what I get with what I think it's best setting ratio quality/weight for Youtube. Note that quality is enought for art like painting. The key value for QP is 27.


    So here is my EPR file to find these settings in Premiere. It must be placed in your folder:

    Windows: C:\Users\UserName\Documents\Adobe\Adobe Media Encoder\12.0\Presets

    Mac: Documents/Adobe/Adobe Media Encoder/12.0/Presets


    Note that you need a GTX 16x0 or a RTX card, otherwise file will be approximately 2.5 times larger (The reason is the availability of B-frames) that is almost the same size than with H264.


    A sample, I get a 207MB file for a 4mn17s video HD1080 at 25fps :

    https://youtu.be/7au1nUmE28Q

    and the original file from Voukoder uploaded to Youtube : https://fromsmash.com/tIhB3eaAMN-ct (available for one week)


    Thanks for your feedback if you think there is still room for improvement

    I investigate a little about nlmeans denoiser. So the same nlmeans developped by Dirk Farin is implemented with FFmpeg and Handbrake.


    There has been an evolution for the version implemented in Handbrake with the support of multithreading. Interestingly enough, the Handbrake source code is available under GNU 2.0 license.

    On the other hand I don't know if there is a release of FFmpeg with a nlmeans multithreading version.

    Maybe Vouk or someone else is able to find this information?


    Dirk Farin's site about nlmeans :

    https://www.dirk-farin.net/projects/nlmeans/index.html


    About specifique release for Handbrake :

    https://forum.handbrake.fr/viewtopic.php?f=4&t=30135

    Hello,


    I just remade tests with the 3 filters of denoising proposed by Voukoder.

    Starting with one minute of HD1080 video at 25 fps, I get the following times with Nvenc :


    HQDN3D filter: 31s

    removegrain filter: 27s

    nlmeans filter: stopped at 3 minutes 35.772ss for 7% of the treatment, i.e. 3071 s for the whole, or an estimated time of 51min 11s


    From a point of result, HQDN3D is satisfactory in particular by decreasing its default parameters a little which remove a little too much detail.

    Nlmeans gives the best visual result but it is unusable due to a rendering of 0.5 fps, in particular because it exploits a single thread.

    There is nlmeans in the Handbrake converter, the processing increases the encoding time but in a more acceptable way since I go up to 12 fps in encoding, this is due to the support of all CPU's threads whose overall load varies from 50 to 75%.


    Currently with support for a single thread, this version of nlmeans which requires a lot of computing power is unusable. May be there is another version of this filter, including the one used by Handbrake?

    Very Interesting and thx urik.


    But hard encoding for H264 at least by Nvidia would already be accessible in the current version or even older versions but without being activated by default :


    - start Premiere Pro

    - CTRL + F12 to display console

    - Activate Debug Database View :

    - search HW to activate hardware encoding by Nvidia with H264 :

    - Restart Premiere Pro

    - do an export, choose H264, there should then be the choice of hard encoding:


    Personally as I use Windows 7 Pro with Premiere Pro 2018 which is the latest supported release. I can actually activate the encoding but it is not selectable for export; Maybe my GTX1660 is not recognized because it came out after.