16 Replies Latest reply on Jul 31, 2009 3:15 PM by ryta1203

    Is ATI Stream dead?


      I'm the author of Pyrit and I wonder If ATI has buried the whole GPGPU thing again?

      The SDK 1.4-beta has not been finalized in months despite many, many obvious glitches and bugs.

      Although Linux 2.6.29 has been out for a whole while now, ATI has not been delivering new drivers that are compatible to this kernel version - effectivly locking out almost all Linux developers that run on less than a few months old kernel versions.

      The OpenCL-functionality should have been on the road in the first half of 2009. Now there is not even a beta.

      There is no word, no roadmap, no guideline from ATI what they are all about


        • Is ATI Stream dead?

          It's possible.

          1. These forums don't get support like they used to.

          2. SDK overdue (according to their schedule)

          3. No\bad tools support (SKA was supposedly scheduled to come out every month but it's been awhile).

          4. OpenCL is off schedule.

          Personally, I don't think so though, I just think AMD is always behind NVIDIA, whether because of talent, finances or simple man hours, I don't know.

            • Is ATI Stream dead?

              Looks like I bought wrong cards after all, and learn obsolete things. But I have been far walking from the right path. I can't go back.

              ~That's my opinion~

                • Is ATI Stream dead?

                  Until now, I have to say we choose a wrong way. I don't know why the development of its software framework is rather slower than its hardware.

                  I think AMD will release the OpenCL SDK finally, but I really doubt its efficiency compared with Nvidia's.

                    • Is ATI Stream dead?

                      It's too bad too.  Ati's 4000 series is really beating Nvidia's 200 series in terms of game performance.  Im primarily a gamer, but I also program in my free time.  Cuda is better than both Brook+ and OpenCL.  From the code on wikipedia, I hope I never have to use OpenCL, regardless of the benefits.

                        • Is ATI Stream dead?

                          OpenCl will most probably take ages before getting new revisions that support new hardware features. That is the worst thing about it I guess. Look at OpenGl.

                          CUDA seems to be the best platform. It is very well documented and gets updated on a regular basis. But ATI has the better hardware and that is why I am here. My feeling is that ATI's coming hardware will be even better than nVidia's future offering, but that is just a feeling :/

                          Choosingn a GPU development platform never been worse.

                          We also need to remember that ATI has the cheaper/faster hardware. Which will probably mean less margins, which means less money to spend on employees that can take care of the documentation and new software revisions.

                          • Is ATI Stream dead?

                            yes, I would really like to use the FireStream 9270 or the Radeon HD 4870x2 in our CFD cluster, but we need OpenCL and X11-free remote access for that, very soon, even if it should be a pre-beta version (just to see that something is on the way).

                              • Is ATI Stream dead?

                                I don't think they plan to drop Brook+, but they are working ******* OpenCL to get it as soon as possible, as nVidia has a pre-release is important to get it fast (probably they're under NDA and trying to get it done for R5870 launch).

                                Coding OpenCL is more complex than Brook+, so for simple computation Brook+ is still interesting, specially if they're planning interoperability and you can use both.

                                I think Brook+ is already quite complete, at least according to the stream programming model and comparing to the original BrookGPU project implementation. There are only a few bugs left and some features like local arrays.