$ | >

OpenCL & CUDA GPU support

It would be nice to have (relatively) easy access to the dedicated GPU's on our Win10 PC's, and OpenCL support for the AMD Drivers & CUDA for NVidia drivers etc.

What I am thinking on specifically (although there are many many uses for it and CUDA support) is hashing functions that are lackluster within windows itself, but benefit immensely within Linux and Unix.

Cryptocurrency mining for example, is known to be much faster on Linux systems than Windows. Do you see any support for OpenCL and/or CUDA in the future?

1,551 votes
Sign in
Check!
(thinking…)
Reset
or sign in with
  • facebook
  • google
    Password icon
    Signed in as (Sign out)

    We’ll send you updates on this idea

    Zachary BoltZachary Bolt shared this idea  ·   ·  Flag idea as inappropriate…  ·  Admin →
    on the backlog  ·  richturnAdminrichturn (Admin, Windows Developer) responded  · 

    It’s on the backlog, but not currently planned.

    We know many of you would like to be able to accelerate your numerical analysis code via CUDA/OpenCL, etc.

    One thing to explore in the meantime is to try running your code in Windows. Windows actually has very powerful GPU-acceleration capabilities and you may well find that much of your existing code may well work already or require trivial effort to port to Windows itself.

    in the meantime, bear with us, and UPVOTE THIS ISSUE if you care (please don’t create another duplicate ask – that doesn’t help us prioritize asks).

    Thanks.

    150 comments

    Sign in
    Check!
    (thinking…)
    Reset
    or sign in with
    • facebook
    • google
      Password icon
      Signed in as (Sign out)
      Submitting...
      • Geoffrey ChurchillGeoffrey Churchill commented  ·   ·  Flag as inappropriate

        I use WSL as a fast, integrated alternative to virtualization. In this era of ML devs and crypto-piggies, native GPU support is an big step toward broader integration and, thus, adoption.

      • Francisco FernandesFrancisco Fernandes commented  ·   ·  Flag as inappropriate

        Yes, we need feedback! So far, WSL is totally useless for me! It cannot even handle the ROS (Robot Operating System) yet! Even with the socket implementation ROS nodes cannot communicate with ROS nodes in other machines! Seriously, WSL is totally useless so far.

      • AwokeKnowingAwokeKnowing commented  ·   ·  Flag as inappropriate

        We really do need feedback! If this is unlikely to happen by the end of the year, a whole group of us will just forget about WSL. Act now, and you'll have much more support from the community than if you hold off until next year.

        This request is similar to "Make IE support JavaScript". Seriously.

      • DawidDawid commented  ·   ·  Flag as inappropriate

        I would also appriciate all this on bash! Running single cryo-EM particle analysis on windows bash - dreams

      • MeteorheadMeteorhead commented  ·   ·  Flag as inappropriate

        Out of professional curiosity (to give us, low-life end-users some perspective): how problematic is it to marshal memory from a Win32 process to a pico process? OpenCL acceleration could totally be solved via a non-conforming libOpenCL.so, one that relays ICD detection over to the Windows ICD loader (OpenCL.dll). The way I imagine such a feature is that once an application invokes the clGetPlatformIDs function, the modified libOpenCL.so would spawn a Win32 process which will invoke the matching function inside OpenCL.dll. All OpenCL handles are type erased structures, libOpenCL.so is free to return whatever pointer that is suitable to match co-existing handles in the Win32 process doing the relay.

        Is it possible to marshal the result of API functions from the Win32 helper process back into an LXCore pico process?

        This way one could be free of implementing GPU driver kernel modules, which are a moving target anyways, but various API versions are fairly static, or at least are far less of a moving target than a vendor driver stack.

      • Non3Non3 commented  ·   ·  Flag as inappropriate

        Upvote, then i can deinstall my dualboot - would be great :)
        I use mostly Hashcat.

      • Anonymous commented  ·   ·  Flag as inappropriate

        Seriously, this feature is going to be in huge demand in the next few years. ML is blowing up and if Microsoft wants a big share of this sector they need to fix this soon. Love the WSL in general, by the way. Great work!

      • AwokeKnowingAwokeKnowing commented  ·   ·  Flag as inappropriate

        I don't want to dualboot. do this. "not planned" == "not thinking straight". Nobody is going to say "oh well I'll just use CNTK". People have needs for linux, and machine learning is a huge one.

      • MikeMike commented  ·   ·  Flag as inappropriate

        Another voice for OpenCL/CUDA support! Bash for Windows was a fantastic step, with GPU support it could become immensely powerful.
        Don't come so close and then fall at the last hurdle, MS!

      • Anonymous commented  ·   ·  Flag as inappropriate

        This is super important, being on Windows is very difficult when working with machine learning. If MS wants to really get more scientists on board the platform there needs to be an easier workflow for apps that require bash and/or Unix tools along with GPU acceleration.

      ← Previous 1 3 4 5 6 7 8

      Feedback and Knowledge Base