Announcement

Collapse
No announcement yet.

Sea Islands Gets AMD R600 LLVM Compute Support

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

  • #11
    Originally posted by pingufunkybeat View Post
    No insult intended
    i mean the fb part

    Comment


    • #12
      Originally posted by pingufunkybeat View Post
      GPGPU is only a drop in the bucket anyway. It's a tool for extremely parallel and extremely heavy computation. Most people don't do extremely heavy computation.

      This is stuff useful for the scientific folks, it won't speed up your facebook or angry birds.
      It's useful for Facebook and Angry Birds. Offloading calculations to the GPGPUs benefits both applications. It's up to the Application to leverage the technologies that will give it more scalability, reduced latencies, etc.

      Comment


      • #13
        Originally posted by kaprikawn View Post
        This is cool and all, but is there any real-world examples of implementations of this? I mean something that people use day-to-day, not something like bitcoin mining.
        SI desktops require it. (I mean, compiling shaders, not the OpenCL stuff)

        Btw, anyone has the idea whether SI is ready for Gnome 3 desktop?
        I tried one month earlier but it was buggy. I switched HD 4600 later and have never tried again.

        Comment


        • #14
          Originally posted by zxy_thf View Post
          SI desktops require it. (I mean, compiling shaders, not the OpenCL stuff)

          Btw, anyone has the idea whether SI is ready for Gnome 3 desktop?
          I tried one month earlier but it was buggy. I switched HD 4600 later and have never tried again.
          Using HD 7850 as my main desktop since Fedora 19 beta (> 6 months), almost everyday, zero bugs. Gnome shell transitions, firefox scrolling, few xonotic levels, all just work from box. What it need to patch to make it "buggy"? Install "most popular distro" ?

          Comment


          • #15
            OpenACC

            OpenCL itself isn't terribly interesting from a general user point of view since it requires developers specifically target code to take advantage of the various implementations which does somewhat limit uptake. OpenACC (possibly accelerated with an OpenCL backend) is much more interesting; with the current specification, the programmer is free to use C, C++ or Fortran, and just requires OpenMP-like compiler directives to add support to code with even the possibility of the compiler auto-parallelising/vectorizing existing code to take advantage of heterogeneous parallel systems.

            Comment


            • #16
              Originally posted by Marc Driftmeyer View Post
              It's useful for Facebook and Angry Birds. Offloading calculations to the GPGPUs benefits both applications. It's up to the Application to leverage the technologies that will give it more scalability, reduced latencies, etc.
              But it's all theoretical right now, isn't it? FB and AB do not leverage OpenCL.

              Comment


              • #17
                Originally posted by storm_st View Post
                Using HD 7850 as my main desktop since Fedora 19 beta (> 6 months), almost everyday, zero bugs. Gnome shell transitions, firefox scrolling, few xonotic levels, all just work from box. What it need to patch to make it "buggy"? Install "most popular distro" ?
                HD 7950 in UNUSABLE even for basic desktop stuff: https://bugs.freedesktop.org/show_bug.cgi?id=69341
                I wanted to downgrade to HD6950 but since they are still quite expensive I'm installing fglrx right now. I hate that shit but it's still better than a crappy desktop
                ## VGA ##
                AMD: X1950XTX, HD3870, HD5870
                Intel: GMA45, HD3000 (Core i5 2500K)

                Comment


                • #18
                  On the list of opencl uses, isn't AMD & LibreOffice going to do some work on the spreadsheet program so that it can use opencl?

                  Comment


                  • #19
                    Blender and OpenCL might be an interesting combination once finished.

                    Comment


                    • #20
                      While not everybody does RAW-development of pictures, I enjoy the speedup when using Darktable :
                      All darktable core functions operate on 4x32-bit floating point pixel buffers, enabling SSE instructions for speedups. It offers GPU acceleration via OpenCL (runtime detection and enabling) and has built-in ICC profile support: sRGB, Adobe RGB, XYZ and linear RGB.

                      Comment

                      Working...
                      X