New Upcoming ATI/AMD GPU's Thread: Leaks, Hopes & Aftermarket GPU's

Discussion in 'Videocards - AMD Radeon' started by OnnA, Jul 9, 2016.

  1. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,349
    Likes Received:
    3,098
    GPU:
    5700XT+AW@240Hz
    As I wrote, pure raytracing can be distributed even across multiple GPUs while they render same frame. And it will not result in loss of efficiency.
    Same goes to CPU raytracing. Only overhead is syncing and telling HW what is on next frame.
    And funny thing, that even nVidia agrees with their old marketing. But as I had to reiterate, key word here is "pure".
    One day we may have that in games. And that's why @Astyanax 's statement which puts emphasis on "never" having chiplets in consumer/gaming space is incorrect.
    (Unless our world comes to end before things get figured out or manufacturing enables chiplets powerful enough to run pure raytracing in real time.)
     
  2. Astyanax

    Astyanax Ancient Guru

    Messages:
    9,131
    Likes Received:
    3,124
    GPU:
    GTX 1080ti
    Last edited: Jan 14, 2021
  3. Astyanax

    Astyanax Ancient Guru

    Messages:
    9,131
    Likes Received:
    3,124
    GPU:
    GTX 1080ti
    "multiple gpu's" is not doing BVH on GPU A and Drawing on GPU B.

    multiple gpu's are doing the work per frame with each card doing the entire work for its assigned frame.
     
  4. Truder

    Truder Maha Guru

    Messages:
    1,345
    Likes Received:
    293
    GPU:
    Sapphire Fury Nitro
    The only thing confirmed about GPU chiplets is a recent patent...

    As for RDNA3, only thing confirmed is that it's going to be on an "advanced node", for now.

    But really, to say consumer GPUs will never have chiplets? That's a very finite statement? AMD engineers are exploring the possibility, I'm sure we can all agree on that? Never say never after all.
     

  5. OnnA

    OnnA Ancient Guru

    Messages:
    12,397
    Likes Received:
    3,074
    GPU:
    Vega XTX LiQuiD
    Unless they do Sole RT Core Chiplet (to handle RT+De-noise)
    So we will have 2 Chips that can do different jobs - it can be done IMhO.
     
  6. Fox2232

    Fox2232 Ancient Guru

    Messages:
    11,349
    Likes Received:
    3,098
    GPU:
    5700XT+AW@240Hz
    https://www.researchgate.net/publication/269273124_Scalable_multi-GPU_cloud_raytracing_with_OpenGL
    Or any other. Like, go for that nVidia's demo. They used those XX GPUs to do real time raytraced look on scene in modeling software. So you could actually see object raytraced as you rotate it and when you stop, you get good image within second.
    If it did not scale, they would not present it for sale.
     
  7. cucaulay malkin

    cucaulay malkin Master Guru

    Messages:
    593
    Likes Received:
    258
    GPU:
    GTX1070 STRIX
    that'd be a waste really and add extra latency
    ray tracing hardware is really a tiny fraction of a gpu die
    the main bottleneck of real time ray tracing is still fp32

    and the main goal of chiplet design is to scale datacenter level products to mainstream.

    [​IMG]

    [​IMG]

    when 600-700mm2 gpus are not enough to power next gen cards then they'll prolly do chiplets,unless they decide bringing back multi gpu is more profitable.
     
    Last edited: Jan 14, 2021

Share This Page