Comments Locked

20 Comments

Back to Article

  • nathanddrews - Monday, November 16, 2015 - link

    Clever play on words or typo:
    "Wrapping things up then, with the Boltzmann Imitative..."

    NVIDIA seems so entrenched, I wish AMD luck on this one!
  • Ryan Smith - Monday, November 16, 2015 - link

    The cleverness is all MS Word's on that one. That's what it corrected my typo as.
  • IanHagen - Monday, November 16, 2015 - link

    Yes, yes, there's no mistake. nVidia must've bribed Microsoft to twist articles in their favour, given that they cannot compete against AMD's brilliant designs fair and square. If only there would be justice on the land!
  • Ryan Smith - Monday, November 16, 2015 - link

    On a semi-related note, I misread the name as the "Bozeman Initiative" the first time around and was disappointed that it was Boltzmann. AMD missed an obvious opportunity for a combination Star Trek + GPU threading joke: "AMD Begins Bozeman Initiative: Acquires NVIDIA Warp Technology"
  • ant6n - Monday, November 16, 2015 - link

    The Bozeman was stuck for 90 years in a time loop, not sure you'd want that association.
  • Wreckage - Monday, November 16, 2015 - link

    CUDA support could also lead to Physx support. This could lead to greater adoption in games which will in turn make games better.
  • VoraciousGorak - Monday, November 16, 2015 - link

    It's not really "CUDA support" as much as it is the ability to easily recompile CUDA code to work with AMD GPUs. Likewise, PhysX is NVIDIA IP, so they're not likely to give it up anytime soon.
  • WhisperingEye - Tuesday, November 17, 2015 - link

    I thought the consoles recieved PhysX support, despite the host GPU.
  • WaitingForNehalem - Monday, November 16, 2015 - link

    It's about time!
  • asmian - Monday, November 16, 2015 - link

    >"Developed core principals" - second slide, "Why Boltzmann"

    OUCH! I hope that wasn't a public presentation slide... The curse of the auto-correct again, perhaps, but it shouldn't get past a PR team.
  • MikhailT - Monday, November 16, 2015 - link

    Why would anyone do this with the legal issue about HIP?

    I fair to see why people should be excited about this? AMD should've presented a better solution for OpenCL and not invest into a solution for a competitor that they have to play catch up with and not only that, they could face legal issues for.
  • Yojimbo - Monday, November 16, 2015 - link

    They have to play catch up either way. Are there really legal issues for writing your own compiler for code written for someone else's compiler?
  • Alexvrb - Monday, November 16, 2015 - link

    OpenCL is an open standard with some serious shortcomings, and poor support on popular Nvidia hardware. Extending this would not solve all their problems in the short term, and anything they add that isn't standard simply would not see much use. What they did was literally the smartest thing they could have done and it involves a three-pronged approach. Continue to support existing and future OpenCL solutions, add extensive support for C++/C to produce CPU and GPU code using a single compiler (HCC), and add support for converting existing CUDA code to run through said HCC *or* NVCC to produce binaries for both major vendors using existing CUDA code.

    If they only did the first or even the first two they'd still be "playing catch up" in your eyes, and they wouldn't be interesting to those currently standardizing on CUDA and Nvidia. Now they're going to get a lot more interest from formerly Nvidia-only outfits.
  • johnpombrio - Monday, November 16, 2015 - link

    Once again, AMD comes out with a massive paper launch and media conference of a non-existent software and programming solution with no dates on when it will be available or how AMD will be able to pay for creating it. Power point slides, slick presenters, and nice graphics are a lot cheaper than actually creating the software that they represent. As with the Fury X launch with its massive AMD hype before launch and its meh performance when it was released, AMD found out how hard it is to actually produce what it said it would in the AMD paper launch of the product years before. Meantime, NVidia and Intel continue to steal away market share without all the press releases and power point slides and media talks. AMD, continue to generate more FUD and smoke and mirrors please. It gives the press something to write about.
  • MrSpadge - Tuesday, November 17, 2015 - link

    Yeah, Intel & nVidia never present their new products and ideas. And never fail to deliver once they show something.. like Tegra, Itanium etc.
  • anubis44 - Tuesday, November 17, 2015 - link

    The Fury X is impressive. The card is designed for DX12, not decrepit DX11, which is all the reviewers used to test the card. In addition, the latest Windows 10 drivers have produced significant performance increases for all AMD cards:

    http://wccftech.com/amd-r9-fury-x-performance-ahea...

    The tables are turning, and AMD is getting stronger, not weaker.
  • D. Lister - Thursday, November 19, 2015 - link

    Fury is designed for an API that would come in full swing, maybe in the next couple of years, when the Fury would already be a generation or two old? Secondly, DX11 has been around for quite a few years, yet DX9 still prevails. The two of these APIs shall continue to exist for a long time right beside DX12.

    Also, wccftech is not really a very good choice as a reference.
  • hammer256 - Monday, November 16, 2015 - link

    Back when Ryan first wrote about GCN I remember thinking that hopefully AMD's software support for programming the GPUs will get better, and maybe get to the point of being as well documented and convenient to use as CUDA. I guess it took them a while...

    In terms of HPC, I think that their effort in this is a lot bigger deal than their next GPU. Hardware is nice and pretty, but it's the software support that allows the hardware to be practical for us ordinary folks to program.
  • Marcelo Viana - Saturday, November 21, 2015 - link

    Well I see no rights of NVidia to sue AMD in any way.
    The AMD tools take a(uncompiled) source code that is a programmer authorship. A code that i use the notepad to write if i wish to, as a small code in c++.
    If AMD use Cuda code in their gpu will be another history, but after all AMD use only programmer authorship code that AMD tools translate to C++ code and translate that C++ code with their own compiler to run in their firepro cards.
    Very smart move from AMD. Lots of ressources, imagine one gpu on node 15 share a memory of the gpu on node 512, through the infiniband, or all the Ram and Vram of the entire system as one unified memory. So easy to program now, finally AMD, finally. The only thing i hope is that their compiler give performance and not just that flexibility and ressources.
  • Marcelo Viana - Saturday, November 21, 2015 - link

    Another thought that i forget.
    So now will be better to program in HIP since the language is almost equals to cuda, somebody post on site can't remember now, but all the commands you change "cuda" to "hip" like:
    cudaMalloc() became hipMalloc()
    cudaMemcpy() became hipMemcpy()
    cudaDeviceSyncronize() became hipDeviceSyncronize()
    and so on.
    the diference is that with HIP i can choose for what gpu i want my code to compile. Where in cuda i have only one.
    One make a gpu render software in CUDA, other make a gpu render software in HIP, but only the second one compile his code to booth vendors AMD and NVidia.
    Very smart move AMD, very smart.

Log in

Don't have an account? Sign up now