NVIDIA Introduces RAPIDS Open-Source GPU-Acceleration Platform

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Oct 11, 2018.

  1. Hilbert Hagedoorn

    Hilbert Hagedoorn Don Vito Corleone Staff Member

    Messages:
    37,081
    Likes Received:
    6,147
    GPU:
    AMD | NVIDIA
    fantaskarsef likes this.
  2. msotirov

    msotirov Active Member

    Messages:
    54
    Likes Received:
    28
    GPU:
    rx 480
    Did I just read Nvidia and open source in the same sentence?
     
    KissSh0t and fragba1t like this.
  3. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    11,287
    Likes Received:
    3,333
    GPU:
    2080Ti @h2o
    I bet it won't be long until people complain about one thing or the other.
     
  4. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,940
    Likes Received:
    2,293
    GPU:
    5700XT+AW@240Hz
    I think more funny is: "today announced" and "with broad adoption" in one sentence.

    "Open Source" has many definitions. Some are pretty surreal. If you asked same question to Torvalds and to Huang regarding vision of Open Source, you would be in for absurd comedy.
     

  5. SetsunaFZero

    SetsunaFZero Active Member

    Messages:
    87
    Likes Received:
    7
    GPU:
    Msi 980Ti 6G
    not quite open as you would think. RAPIDS only works with nvidia CUDA
     
  6. pharma

    pharma Maha Guru

    Messages:
    1,184
    Likes Received:
    159
    GPU:
    Asus Strix GTX 1080
    RAPIDS was designed by Nvidia and utilizes acceleration aspects of CUDA. Open Source code is available so other IHV's can modify and offer their own acceleration solutions within the RAPID framework. Similar to Vulkan where different IHV's offer their own code for features to run on their GPU's.
     
  7. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    11,287
    Likes Received:
    3,333
    GPU:
    2080Ti @h2o
    I just have to ask: Why should Nvidia spend engineering hours (money) on programming for GCN specifically? That would not make much sense, would it? Especially not from a business perspective.
     
  8. kakiharaFRS

    kakiharaFRS Master Guru

    Messages:
    275
    Likes Received:
    55
    GPU:
    MSI Gaming X 1080ti
    don't tell me they are giving another use to gaming videocards ><
    just create other hardware cards with the same gpu inside to split the markets and yes you could sell at even more expensive prices to professionnals (if you do not work in the industry you'd be shocked to know the prices for electronic hardware in it we aren't talking in "ryzen" prices but in "same as a brand new car" prices)
    you only care about the 1st sell anyway not the resell market so please just stop :(
     
  9. SetsunaFZero

    SetsunaFZero Active Member

    Messages:
    87
    Likes Received:
    7
    GPU:
    Msi 980Ti 6G
    true but AMD or third party could adopt it to GCN. Nv probably wouldn't like that.
    I wouldn't be surprised it they added some AMD detection code to shut RAPIDS off just like Physx.
     
  10. fantaskarsef

    fantaskarsef Ancient Guru

    Messages:
    11,287
    Likes Received:
    3,333
    GPU:
    2080Ti @h2o
    You know what open source means? They can look at the code and adopt it, right? If not please forgive me, it's how I understood it at least.
     

  11. Fox2232

    Fox2232 Ancient Guru

    Messages:
    9,940
    Likes Received:
    2,293
    GPU:
    5700XT+AW@240Hz
    Well, that's for open source, and very true. Yet, it is unlikely to happen as it did not till now. Reason is simple, It runs on CUDA and is tied to it. To get it working, you make fork which would stop using CUDA and put it to OpenCL/DirectCompute. (Lot of work and always behind.)
    CUDA itself as base thing is not open source, therefore not implementable for AMD's or intel's HW.

    In other words Open Source application which is using CUDA as base does not really provide usual advantages of Open Source.

    If there was Compute Abstraction Layer in which you could 1 click switch between CUDA/OpenCL/DirectCompute and get compiled library, then Open Source project like this could benefit from being Open Source in terms of availability for wide use.

    You can think of this in way that having Open Source CUDA based software is same as having Open Source game which uses direct instruction set for AMD's graphics cards (skipping DirectX) and therefore not being available for anyone else.
    (In that case one would have to take that open source and translate it to intel/nVidia. Or simply use abstraction layer called DirectX.)
     
    Embra and fantaskarsef like this.

Share This Page