Announcement

Collapse
No announcement yet.

NVIDIA Puts Out A Major Beta Linux Driver Update

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

  • #31
    Originally posted by Silent Storm View Post
    Yes, they have. There's even a small benchmark.

    OTOH, OpenCL is a standard and CUDA is just a vendor specific API. I'd also prefer OpenCL even it is a bit harder to code.
    I also believe nVidia offered CUDA to AMD as well.

    Comment


    • #32
      Originally posted by DeepDayze View Post
      I also believe nVidia offered CUDA to AMD as well.
      Only if AMD gave NVidia all their hardware specifications, as i recall. Not going to happen.

      Comment


      • #33
        Originally posted by tball View Post
        No I think we have to see OpenCL added to mplayer etc. CUDE only works on nvidia gfx's, but OpenCL works on almost all vendors.
        What are you getting at? OpenCL isn't that useful for video decoding.

        Comment


        • #34
          Originally posted by smitty3268 View Post
          Only if AMD gave NVidia all their hardware specifications, as i recall. Not going to happen.
          Can you back that up?

          Comment


          • #35
            Originally posted by brent View Post
            What are you getting at? OpenCL isn't that useful for video decoding.
            I wouldn't say that, it can be useful just not as efficient as dedicated hardware. CoreAVC has a successful Cuda decoder and there is also a Cuda Dirac decoder as well. OpenCL would be just as suitable for a GPGPU framework to get those tasks done. The real power however would be however the flexibility in post processing and/or encoding.

            Comment


            • #36
              CoreAVC merely uses nvcuvid, to get around the artifical limitations of DXVA, but that's it. Never heard about the Dirac decoder though.

              I don't doubt that partial acceleration with a shader-based video decoder is possible, but I very much doubt that it will be efficient enough to be interesting in practice. Also, it's not trivial at all to implement.

              Just look at how very mediocre offload was with NVidia's/AMD's early attempts at video decoding that only accelerated MC and deblocking. You won't get more acceleration than that with shader-based decoding. Heck, they didn't put all that extra logic for video decoding into newer chips for fun.

              Comment


              • #37
                What is the premiere VDPAU-capable media player? I've been using a mix of VLC and SMplayer for a few months, but neither seems to work acceptably.

                VLC works the best, but the OSD control in fullscreen is all black on my system. That's just a minor cosmetic issue. The bigger issue is that VLC doesn't support VDPAU (at least the current Fedora version).
                SMplayer with VDPAU is great for my system, buts it has lots of bugs for me. If I resize the video (i.e. fullscreen) or pause/then play the video, the video freezes but audio plays correctly. I have to seek to get it working again.

                I assume that my troubles with SMplayer are mostly pebkac. Anyone know a workaround, or a better application?

                Comment


                • #38
                  Originally posted by brent View Post
                  What are you getting at? OpenCL isn't that useful for video decoding.
                  I hate to disagree but, ATI had an OpenGL demo called VideoShader demo that postprocessed the video of your choice on the fly and it was a big hit with 9500/9700 cards. They were also able to decode and accelerate H.264 videos on a 9600XT, even on Linux (I have witnessed negligible CPU loads while decoding some HD content using XVideo acceleration).

                  If they were able to do this with pure OpenGL, I think it's possible to decode video on a chip that can manipulate matrices just for fun. In fact, video decoding is just altering frames (which are matrices) according to a original template (keyframe) using some math-magic.

                  It can be done and the development process shall be fun, I presume.

                  Comment


                  • #39
                    Originally posted by jbrown96 View Post
                    What is the premiere VDPAU-capable media player?
                    By far the best implementation so far is XBMC (MythTV as well). SMplayer works well as well, at least on openSUSE and your problem doesn't seem replicable on it.

                    Comment


                    • #40
                      Originally posted by deanjo View Post
                      Can you back that up?
                      Hmm, I guess not. I got that from another poster and no one called him out on it, so i figured it was legit. Now that i've looked into it, i'm not sure where it came from and whether it was true or not. I believe what he was saying was that nvidia was offering to build support for amd chips into their compiler, which obviously would require giving them knowledge of how to program it.

                      Anyway, i'm not sure why anyone would choose CUDA for something cross platform right now anyway. I thought that was the whole point of OpenCL, to create a language that was compatible between different hardware. Why would AMD suddenly create a 2nd language, one that it's competitor controls and could modify to make their hardware look bad.

                      Comment


                      • #41
                        Originally posted by smitty3268 View Post
                        Anyway, i'm not sure why anyone would choose CUDA for something cross platform right now anyway. I thought that was the whole point of OpenCL, to create a language that was compatible between different hardware.
                        Well at this time there are still compelling reasons. Cuda is relatively mature and openCL still has some teething pains (such as lack of language bindings). Even with openCL you still need device specific optimizations to make the most out of it, just because something compiles with openCL doesn't mean that it's going to be optimized for all openCL devices. Also the documentation and resources and tools on Cuda are vast compared to any other GPGPU solution out there. There is also the fact that you have to run blobs at the moment to take advantage of any of the GPGPU solutions at present. I'm sure if you searched the forums you can find a few posts out there showing the difference's between openCL and Cuda (such as memory management).

                        Those are just a few reasons, there are many others but it basically boils down to a case of a mature solution vs a solution in progress.

                        Comment

                        Working...
                        X