Announcement

Collapse
No announcement yet.

Gdev Open-Source CUDA Runtime Is Still Running

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

  • Gdev Open-Source CUDA Runtime Is Still Running

    Phoronix: Gdev Open-Source CUDA Runtime Is Still Running

    Gdev, the GPGPU run-time and resource management engines that provides an open-source NVIDIA CUDA run-time, is still being worked on at the University of California Santa Cruz in conjunction with PathScale...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Why use CUDA instead of OpenGL?

    Why use CUDA instead of OpenGL?
    It seems retarded.

    Then you stuck with Nvidia forever and cant change to another vendor.
    Maybe Nvidia goes under in some years, then what do you do now with your legacy CUDA stuff?

    Why not just OpenGL instead which is a standard that works everywhere on everything?

    Comment


    • #3
      Originally posted by uid313 View Post
      Why use CUDA instead of OpenGL?
      Don't you mean OpenCL? You can do GPGPU programming with OpenGL (transform feedback with geometry shaders etc.), but it's less intuitive.

      But either way, nVidia's CUDA developer tools are absolutely fantastic, and there's more books/resources available to developers.

      Comment


      • #4
        Originally posted by cbamber85 View Post
        Don't you mean OpenCL? You can do GPGPU programming with OpenGL (transform feedback with geometry shaders etc.), but it's less intuitive.

        But either way, nVidia's CUDA developer tools are absolutely fantastic, and there's more books/resources available to developers.
        Sorry, I posted in the morning and was tired.
        Yes, I obviously meant OpenCL, not OpenGL.

        I don't know CUDA or OpenCL but CUDA being vendor-specific and OpenGL being an open standard, it seems to be that OpenCL would be the obvious choice.
        IMHO, using a vendor-specific technology is just idiotic.

        Comment


        • #5
          in the proffessional world, using vendor specific tools is just the way it is.

          its sad, its stupid, yes, but as an example, look at opengl extentions. if a feature is a good idea, and makes sense, the opengl board would design an implementation and add it to the next opengl spec, at least thats how your would think it would work, the reality is that the hardware vendors will (or more would, i think it was used a lot more in the past) get feedback from some of their top clients (think people who use the cad programs and major model renderers or movie developers, you know, the same people nvidias blob is said to accomidate for) and the would create a often quick and dirty extension for them to use.

          often these high end users like companies dont care about the standardization of the tools they use, they just want the quick and easy solution thats right i front of them, or, more truthfully, the guy making the decision isnt the developer, its the bisness exec who has no fscking clue about the stuff hes suspose to be in charge of and just bends over for some salesmans pitch about speed and efficiancy and proficiancy and whatever buzzword you can come up with that makes you more pissed off.

          Comment


          • #6
            I missed this when it was posted. Development continued for a couple of more years but seems to have stalled since 10 Nov 2014.

            I've been taking a look at GCC Offloading, currently implementations are limited to the NVPTX and Intel MIC backends which really limits interest and isn't really in the spirit of GNU or F/OSS.

            So, I was thinking at looking whether it would be possible to create a Gdev GCC Offload backend.

            Further ideas occur to me, perhaps look at using the Intel i965 compiler in Mesa to be able to target Intel HD Graphics hardware, and/or adapt the HSA AMDGPU compiler from the GCC HSA branch to target non-HSA AMD GPUs?

            Crazy ideas?

            Comment

            Working...
            X