Announcement

Collapse
No announcement yet.

Google Working On NVIDIA CUDA Support Via LLVM?

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

  • rabcor
    replied
    Originally posted by Pajn View Post
    Google is pretty interested in machine learning which it uses for Google Now and there self-driving cars but are slowly extending past that.
    Machine learning is an area I would expect get a big performance jump on the GPU, neural networks get better the more synapses you have.
    This is it, I think Google is cozying up to Nvidia for machine learning. Watch this video, and that video:



    You'll understand why Google would be doing this.

    Leave a comment:


  • Sdar
    replied
    Originally posted by wizard69 View Post
    Seriously why would Google be investing in a proprietary GPU compute environment! I could see a benefit in a generic GPU compute solution like OpenCL but it just strikes me as poor management to invest in CUDA. That would apply to any company really.
    Have you seen the pascal presentation on the GTC 2014?
    They just outperformed google brain (1.000 servers, 16,000 cores) with just 12 gpus.

    P.S. Companies are struggling to make opencl work for full featured rendering engines, they have to split the kernel and do lots of workarounds just to be able to compile, so it seems that amd Opencl compiler is not mature enough yet.
    Last edited by Sdar; 22 April 2015, 07:11 AM.

    Leave a comment:


  • wizard69
    replied
    Isn't this a big waste on Googles part?

    Seriously why would Google be investing in a proprietary GPU compute environment! I could see a benefit in a generic GPU compute solution like OpenCL but it just strikes me as poor management to invest in CUDA. That would apply to any company really.

    Leave a comment:


  • Pajn
    replied
    Google is pretty interested in machine learning which it uses for Google Now and there self-driving cars but are slowly extending past that.
    Machine learning is an area I would expect get a big performance jump on the GPU, neural networks get better the more synapses you have.

    Leave a comment:


  • Danny3
    replied
    What's the point

    So what's the point developing closed source software?

    Leave a comment:


  • phoronix
    started a topic Google Working On NVIDIA CUDA Support Via LLVM?

    Google Working On NVIDIA CUDA Support Via LLVM?

    Phoronix: Google Working On NVIDIA CUDA Support Via LLVM?

    For whatever reason, there's Google developers working on CUDA improvements within the LLVM/Clang compiler...

    http://www.phoronix.com/scan.php?pag...M-CUDA-Support
Working...
X