Loading...
> >

Run CUDA on AMD?... OTOY thinks so...

McGyverMcGyver Posts: 7,085
edited May 2016 in The Commons

Terrible thread title, but I've done worse... Anyhooz... I was looking up one thing and came across this:

http://www.bit-tech.net/news/hardware/2016/03/11/otoy-cuda-compiler/1

Which sounds kinda cool... But it referees to this article:

http://venturebeat.com/2016/03/09/otoy-breakthrough-lets-game-developers-run-the-best-graphics-software-across-platforms/

Judging from some of the wording which sounds kinda short on info and long on buzz... (no benchmarks? comparisons?)... Not that the idea sounds impossible or anything it's just like more real info could have been added since the article seems more of a blurb for investors than info for tech-geeks or the folks who'll actually use it.

Would it work as smoothly as we would hope... who knows... Will it be soon or SOON™... who knows... How much would it cost? Are there limitations? Will what are the side effects when used with prescription medications?... 

Whatever the case... I couldn't find a previous thread about this here and don't recall one, but I thought since many people have AMD cards, AMD cards are (slightly) cheaper and new Macs only use AMD, this might be worth knowing and keeping an eyeball on.

And if this already has a thread, go ahead and deleted it or decorate it with lol cat, red panda and octopus pictures.

Post edited by McGyver on

Comments

  • StratDragonStratDragon Posts: 3,273

    this is exciting and when the smoke clears from the CEO's of both companies saying their next gen products are 1,100 times more powerful than the Sun power lifting to Billy Joels epically mind blowing spinal-tap-ian venture into heavy metal then I would love to see where this goes (the tech, not the song, I have no idea where that thing is going)

    Sincerely,
    J. Pandapus Brinkly 

     

     

     

    Unknown.jpeg
    284 x 177 - 10K
  • ANGELREAPER1972ANGELREAPER1972 Posts: 4,555

    well says itll be able to run stuff but whether or not it'll be as high quality is another thing kinda sounded like there was limitations in the quality 

  • DustRiderDustRider Posts: 2,880

    One of the features for Octane 3 has been to include the use of non Nvidia cards.  Based on what Urbach said, it sounds like they ran into too many issue with OpenCL where it didn't support the full Octane feature set, so they went to plan "B". Keep in mind that Otoy developed Octane without using the path tracing (rendering) sdk from Nvidia, so they know Cuda extremely well.

    Very interesting news. It will be fun to see what Nvidias response to this will be, I'm sure they can't be very happy about it.

  • MistaraMistara Posts: 38,675

    openCL and openGL are 2 different things?

     

  • jurajura Posts: 50
    edited May 2016

    Hi there

    Have look on GPUOpen,its API which will allows developers to run CUDA on AMD,I just think they're used cross compiler for CUDA,in theory this can be possible to run any CUDA application on AMD

    Here is link http://gpuopen.com/compute-product/hip-convert-cuda-to-portable-c-code/

    Over on Otoy,there has been question regarding performance on AMD cards and they're confirmed there is no performance penalty on AMD

    Hope this helps

    Thanks,Jura

    Post edited by jura on
  • mtl1mtl1 Posts: 1,508
    jura said:

    Hi there

    Have look on GPUOpen,its API which will allows developers to run CUDA on AMD,I just think they're used cross compiler for CUDA,in theory this can be possible to run any CUDA application on AMD

    Here is link http://gpuopen.com/compute-product/hip-convert-cuda-to-portable-c-code/

    Over on Otoy,there has been question regarding performance on AMD cards and they're confirmed there is no performance penalty on AMD

    Hope this helps

    Thanks,Jura

    And just to emphasize that this is a compile-time solution and not a runtime solution. Ideally, the renderer should point to a generic API which then calls whichever graphics card is installed. However, iray is most likely calling CUDA-specific instructions at the driver level so it'll be quite some time before we'll see any kind of cross-platform solution...

  • AlienRendersAlienRenders Posts: 794
    edited May 2016

    "openCL and openGL are 2 different things?"

    OpenGL (cross-platform) and DirectX (Windows only) are for displaying geometry right away. OpenCL (cross-GPU) and CUDA (nVidia only) are for GPGPU programming. So while iRay does display geometry, it uses custom programming to create the final result using CUDA. LuxRender uses OpenCL which allows both AMD and nVidia video cards. I've tried it with both video cards at the same time. It's kinda cool. But very diffucult to setup Lux to actually run correctly. Their CPU renderer is quite good though.

    Also, FYI, Vulkan is the next generation alternative to OpenGL that is supposed to be the equivalent to DirectX 12. These new versions have better support for multithreading amongst other things. The big buzz is the ability to issue MANY more draw calls per second. We're talking several orders of magnitude more.

    Post edited by AlienRenders on
  • mjc1016mjc1016 Posts: 15,001
    AlienRenders said:
     

    Also, FYI, Vulkan is the next generation alternative to OpenGL that is supposed to be the equivalent to DirectX 12. These new versions have better support for multithreading amongst other things. The big buzz is the ability to issue MANY more draw calls per second. We're talking several orders of magnitude more.

    Vulkan is like OpenGL on steroids after a couple of dozen extra large Starbucks...or at least on paper it is.

  • JCThomasJCThomas Posts: 254

    I'd been following this closely since Octane 3 was announce and saw that original story in Venture Beat. It sounded like a really big deal, especially for octane users, because AMD cards are cheaper and OTOY claims the performance is the same. If the performance is indeed the same, then a super octane box could be built on a budget with some R9 390s or 390Xs. But not that we've seen the specs on Pascal, this seems less interesting. 8 GB VRAM on the 1070 and Titan X-like performance for not much more than 970 prices is too good to justify possible conflicts running CUDA on AMD cards. Of course, we have to wait and see what the performance is really like, but the specs make the claims mostly believable in my opinion. I'd also be willing to bet that Iray will really benefit from pascal.

    I agree, it will be interesting to see how NVidia responds.

  • Kendall SearsKendall Sears Posts: 2,995

    The kicker is going to be what nVidia Compute Level is supported by the converter.

    Kendall

Sign In or Register to comment.