Announcement

Collapse
No announcement yet.

NVIDIA Contributing Tegra NVDEC Support To Linux 5.16

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • NVIDIA Contributing Tegra NVDEC Support To Linux 5.16

    Phoronix: NVIDIA Contributing Tegra NVDEC Support To Linux 5.16

    The Tegra DRM driver changes were sent out on Friday of the new material destined for Linux 5.16. Notable this time around is NVIDIA's NVDEC driver being included...

    https://www.phoronix.com/scan.php?pa...nux-5.16-NVDEC

  • #2
    Since they implemented dmabuf support in their proprietary driver, they should also contribute something to make VAAPI in Firefox work on Nvidia (be it some kind of wrapper or whatever).
    Meanwhile, I've compiled mpv with ancient GLX backend to have VDPAU support that allows better downclocking than NVDEC. I'm really not interested in NVDEC when it wastes lots of electric energy for nothing. They should really dump that clocking behavior or limit it to NVENC...
    Last edited by aufkrawall; 10 October 2021, 07:22 AM.

    Comment


    • #3
      Where is the big open source thing, nvidia announced back then?

      Comment


      • #4
        Originally posted by baka0815 View Post
        Where is the big open source thing, nvidia announced back then?
        We might be seeing GBM enabled driver later this month. Could also be interesting what they'll ship on top of that, perhaps Xorg modesetting driver support etc. (just wild guessing).

        Comment


        • #5
          Originally posted by baka0815 View Post
          Where is the big open source thing, nvidia announced back then?
          I think nVidia has been scared away by the steep requirements & standards imposed by the upstream Linux kernel community.

          Plus having reworked their binary blob to make use of GBM further decreased any incentive to open-source the kernel-level part of their driver.

          Therefore, as long as nVidia can get away with it while doing just fine at the same time, things probably are not going to change on their side for the foreseeable future...

          Comment


          • #6
            Well, the fact that they finally stopped trying to make their own VDPAU happen, and adopted VA-API as the industry standard for accelerated video playback, that's progress.

            Comment


            • #7
              Originally posted by SteamPunker View Post
              Well, the fact that they finally stopped trying to make their own VDPAU happen, and adopted VA-API as the industry standard for accelerated video playback, that's progress.
              That's incorrect. They still back VDPAU, but are offering VA-API for those that want it, as it's preferred by many developers. Their current approach is NVDEC/NVENC, which was developed as a multi-platform approach with GPU shaders helping out, but after introducing it as CUVID and then forcing it on the software ecosystem by killing off VDPAU development, the software devs who did the media players and such refused to implement it, because they had already done VDPAU, so Nvidia finally relented and resurrected VDPAU, while simultaneously trying to push NVDEC/NVENC as an eventual replacement.

              Comment


              • #8
                Originally posted by TheLexMachine View Post

                That's incorrect. They still back VDPAU, but are offering VA-API for those that want it, as it's preferred by many developers. Their current approach is NVDEC/NVENC, which was developed as a multi-platform approach with GPU shaders helping out, but after introducing it as CUVID and then forcing it on the software ecosystem by killing off VDPAU development, the software devs who did the media players and such refused to implement it, because they had already done VDPAU, so Nvidia finally relented and resurrected VDPAU, while simultaneously trying to push NVDEC/NVENC as an eventual replacement.
                Lol, they actually named it CUVID? That name's a lot funnier now, maybe they'll give it a few version bumps? CUVID 19.0?

                Comment


                • #9
                  Yes, the name for it was CUVID (CUDA VIDEO). It didn't take because of the VDPAU issue and the glaring problem that using CUVID was essentially running a card in gaming mode and eating up power/creating heat, so they rebranded it with the newer GeForce GPUs as NVidia Decode/NVidia Encode, once they implemented newer and more efficient processing cores and video decoding blocks to use.

                  Comment


                  • #10
                    Interesting, why did NVidia choose to go through DRM and require yet another userspace library for talking VA-API? Couldn't they have used v4l2 request API like e.g. sunxi-cedrus?

                    Comment

                    Working...
                    X