NV40 architecture and multichip GPUs

Discussion in 'Architecture and Products' started by nAo, Apr 3, 2005.

  1. _xxx_

    Banned

    Joined:
    Aug 3, 2004
    Messages:
    5,008
    Likes Received:
    86
    Location:
    Stuttgart, Germany
    ...london-boy, for "Oh, I'm really not that horny at all"? :twisted:
     
  2. Simon F

    Simon F Tea maker
    Moderator Veteran

    Joined:
    Feb 8, 2002
    Messages:
    4,563
    Likes Received:
    171
    Location:
    In the Island of Sodor, where the steam trains lie
    Never mind the complexity of the board design, it's the chip design that you need to worry about.

    I've been trying to do a search of the web for details but, IIRC, the I/O pads and associated logic you need to drive a signal out of the chip, are enormous compared to the usual internals.
     
  3. _xxx_

    Banned

    Joined:
    Aug 3, 2004
    Messages:
    5,008
    Likes Received:
    86
    Location:
    Stuttgart, Germany
    I never heard that was that big of a problem, but what do I know. Let's just add that to the "contra" list :)

    It might still pay off in some specific cases, thanks to better yields and possible scalability. We'll see.
     
  4. Simon F

    Simon F Tea maker
    Moderator Veteran

    Joined:
    Feb 8, 2002
    Messages:
    4,563
    Likes Received:
    171
    Location:
    In the Island of Sodor, where the steam trains lie
    Well, I could be wrong, as I am primarily a software engineer.
     
  5. Unknown Soldier

    Veteran

    Joined:
    Jul 28, 2002
    Messages:
    4,047
    Likes Received:
    1,670
    Dave Baumann - No!

    US
     
  6. Joe DeFuria

    Legend

    Joined:
    Feb 6, 2002
    Messages:
    5,994
    Likes Received:
    71
    Talking about the "pros and cons" of "single chip" vs. "multi chip" is too broad. What we are really theorizing here is more like:

    A) Large, single unified architecture chip (Big R600) vs. disparate multi-chip (smaller nVidia 'chipsets')

    There are different factors related to the above, compared to say:

    B) Large, single chip (Example: R300), vs. am "indentical multichip" (theoretical RV350 SLI)

    Interestingly enough, we already have a case study for A. 3dfx Voodoo2 vs. nVidia TNT-2. One might say that the Voodoo2 had clear early timing ( and scalability) advantages, but the TnT-2 had longevity advantages. (The cost of Tnt-2 scaled better over time)

    We also have to keep in mind that if ATI goes with a "unified" architecture, this says nothing about single or multi-chip plans in and of itself. ATI might just as well go with smaller multichip solutions, just as we are threorizing nVidia might. The difference would be that all of ATI's chips in their solution would likely be "identicial" chips (multiple "unified" chips), vs. in nVidia's case, they would be disparate. (One "vertex" chip, and one "fragment" chip, for example.)

    I would assume there are unique challenges that go with either type of approach.
     
  7. _xxx_

    Banned

    Joined:
    Aug 3, 2004
    Messages:
    5,008
    Likes Received:
    86
    Location:
    Stuttgart, Germany
    But that's because TNT had more fetures. It was essentially a next-gen chip to V2 (32-bit color etc.) and was 2d/3d combo and thus cheaper than a V2 + nice 2d card. I think it's a bad example.
     
  8. Joe DeFuria

    Legend

    Joined:
    Feb 6, 2002
    Messages:
    5,994
    Likes Received:
    71
    Yes and no. TNT was not a superset of Voodoo2.

    We're never going to have a perfect example, but the bottom line is, Voodoo2 and TNTs feature set were largely comparable. Including the 2 TMU to 1 Pixel Pipeline architecture.
     
  9. AndrewM

    Newcomer

    Joined:
    May 28, 2003
    Messages:
    219
    Likes Received:
    2
    Location:
    Brisbane, QLD, Australia
    The TNT had some (for the time) decent texture combiner hardware, which then became Register Combiners for the Geforce, and extended to make "pixel shaders" for Geforce3, and also ended up in GeforceFX.

    At the time, it was a fair bit more than what the V2 could do, which hadnt changed from the V1.
     
  10. nobody

    Newcomer

    Joined:
    Sep 3, 2004
    Messages:
    110
    Likes Received:
    0
    Location:
    Germany

    The TNT has 2 pixel pipelines with one TMU per pipeline.
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...