Announcement

Collapse
No announcement yet.

A CUDA Back-End For Intel's Open-Source Driver?

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

  • A CUDA Back-End For Intel's Open-Source Driver?

    Phoronix: A CUDA Back-End For Intel's Open-Source Driver?

    While there is the "Clover" branch of Mesa started by Zack Rusin for providing an OpenCL state tracker that can be used by Gallium3D hardware drivers, it hasn't yet amounted to much. The OpenCL state tracker is not yet working, hasn't been touched in months, and has yet to be integrated in the mainline Mesa code-base. However, as another GPGPU alternative, it looks like a CUDA back-end that's specific to Intel's open-source driver may end up being worked on...

    http://www.phoronix.com/vr.php?view=ODM2Mg

  • #2
    Seriously, Cuda? On Intel IGP's? Am I the only one who thinks this is one of the most stupid ways of burning recources that could be used a lot better somewhere else?

    I mean; hardly anyone uses Cuda and those who do probably need a lot more then a miserable Intel IGP will offer. So who is going to use this? What is the point of making this?

    Comment


    • #3
      Just another proof that the graphic stack development for linux is, generally speaking, clueless.

      Comment


      • #4
        I think you two are clueless. This is a group of researchers interested in experimenting with GPGPU backends. They're probably not interested in what you're interested in. They're interested in IGPs specifically because they feel that CPU-GPU integration is relevant to their research. Nobody said they were releasing anything or contributing anything to the Intel driver...

        It's not brain surgery here, the guy's mail to the list is only 3 paragraphs long.

        Comment


        • #5
          don't waste your time on intel or nvidia get that stuff: http://www.theregister.co.uk/2010/06...pu//print.html

          Passiv cooled :-)

          Comment


          • #6
            I don't get it.

            If they are really that intersted into CPU/GPU stuff why strive for CUDA that was designed for GPU only?

            I mean hack, OpenCL was at least designed with CPU in mind.
            Yeah I know that to get very good performance you still have to write code specifically for CPU or certain types of GPUs, but still why CUDA?

            Comment


            • #7
              Why? Because they can and any knowledge can be usefull for noveau. Next.

              Comment


              • #8
                Originally posted by V!NCENT View Post
                Why? Because they can and any knowledge can be usefull for noveau. Next.
                Wow. Great argument.

                So people working on Intel drivers are supposed to get the knowledge for an driver for NVidia cards?

                Comment


                • #9
                  This news is actually pretty cool.

                  Comment


                  • #10
                    Originally posted by marek View Post
                    This news is actually pretty cool.
                    Yeah, it's pretty cool, but I'd also like to see them put their efforts into finishing off the OpenCL state tracker.

                    Although, if they manage to get a working CUDA backend for the intel driver, then someone might be able to do like Nvidia does and provide an OpenCL -> CUDA wrapper/translator, which might be portable to other GPU architectures in the open source stack.

                    Comment


                    • #11
                      The way I understand it, they are interested in GPGPU research, not in implementing some state trackers, and their contribution to the open source stack is just a side effect of this.

                      There is no one wasting resources except for the volunteer contributors, who might be wasting their time that can be spent elsewhere, you know, by actually making some money.

                      Comment


                      • #12
                        Originally posted by marek View Post
                        The way I understand it, they are interested in GPGPU research, not in implementing some state trackers, and their contribution to the open source stack is just a side effect of this.

                        There is no one wasting resources except for the volunteer contributors, who might be wasting their time that can be spent elsewhere, you know, by actually making some money.
                        Wait, you're saying that I can't reserve the right to demand that other people donate their free time (or research time) working on projects that I want them to work on? NOO!!!!

                        In my case it's more that I am hoping that the OpenCL state tracker gets finished so that my proposed thesis project is usable by the gallium infrastructure...

                        I'm currently thinking about proposing an OpenCL port of the VP8 codec, which would mean that any OpenCL-capable system could accelerate decode of this new format. Of course, if I don't get any advisors to bite on that, I could try to directly do a VP8->Gallium state tracker... or fall back to something work-related... or try to finish the OpenCL tracker myself.

                        Comment


                        • #13
                          Originally posted by Qaridarium View Post
                          don't waste your time on intel or nvidia get that stuff: http://www.theregister.co.uk/2010/06...pu//print.html

                          Passiv cooled :-)
                          Hah, what a joke, you really think you could rely on Fglrx to do scientific calculations? I couldn't even manage to get compiz to run on it good enough!

                          Hardware, no matter how good it is designed, without brilliant software, is a pile of electric trash, period.

                          Comment


                          • #14
                            Originally posted by FunkyRider View Post
                            Hah, what a joke, you really think you could rely on Fglrx to do scientific calculations? I couldn't even manage to get compiz to run on it good enough!

                            Hardware, no matter how good it is designed, without brilliant software, is a pile of electric trash, period.
                            Just because you can't do it does not mean that others are as incompetent as you are.

                            I can use OpenCL with ATI cards and AMD (and Intel via AMD) CPUs as I can with Nvidia GPUs. So don't make things look bad while they aren't.

                            Comment


                            • #15
                              Originally posted by mat69 View Post
                              Just because you can't do it does not mean that others are as incompetent as you are.

                              I can use OpenCL with ATI cards and AMD (and Intel via AMD) CPUs as I can with Nvidia GPUs. So don't make things look bad while they aren't.
                              They're comparing the results with desktop use for a driver that's intrinsically written to support Workstations doing things like CAD and GPGPU stuff for compute clusters and by happy happenstance does more. I just wish the "more" was more robust than NVidia's "more" because they're doing their drivers for the same reasons and just have a less stringent GL stack than AMD has right at the moment.

                              It should be noted that many of the "issues" are less issues and more due to things like VSYNC being off by default on Linux, faintly (or even RADICALLY so in some cases...) off GL code, and the like.

                              Comment

                              Working...
                              X