Nvidia Volta Speculation Thread

Discussion in 'Architecture and Products' started by DSC, Mar 19, 2013.

Tags:
  1. DSC

    DSC
    Banned

    Joined:
    Jul 12, 2003
    Messages:
    689
    Likes Received:
    3
  2. Bob

    Bob
    Regular Subscriber

    Joined:
    Apr 22, 2004
    Messages:
    424
    Likes Received:
    47
    That was quick...
     
  3. phenix

    Regular

    Joined:
    Feb 22, 2003
    Messages:
    620
    Likes Received:
    1
    Location:
    Cambridge, MA
    1 Tbyte/s memory badwith with stacked DRAM (Titan has 300 GByte/s)
     
  4. UniversalTruth

    Veteran

    Joined:
    Sep 5, 2010
    Messages:
    1,747
    Likes Received:
    22
    What year should we be expecting this for? 2020? :no:
     
  5. phenix

    Regular

    Joined:
    Feb 22, 2003
    Messages:
    620
    Likes Received:
    1
    Location:
    Cambridge, MA

    Next high performance process node succeeding TSMC 20nm. 2016?
     
    #5 phenix, Mar 19, 2013
    Last edited by a moderator: Mar 19, 2013
  6. 3dilettante

    Legend Alpha

    Joined:
    Sep 15, 2003
    Messages:
    8,019
    Likes Received:
    2,569
    Location:
    Well within 3d

    There are enough interested parties for stacked DRAM that I think something will come up before then.

    One key thing is that making Volta's marketing bullet point stacked DRAM is that this can be relatively agnostic to the GPU architecture it is attached to. Delays in architectural design would be one possible source of schedule slippage that won't be a barrier to releasing something with stacked DRAM and calling it Volta.

    Every other challenge to implementing the solution would remain, but everyone wants the bandwidth and power savings. They are going to be hard-pressed to get it any other way.
     
    #6 3dilettante, Mar 19, 2013
    Last edited by a moderator: Mar 19, 2013
  7. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,004
    Likes Received:
    109
    Does this mean that all memory is stacked (so not just like a big cache, but actually ditching "external" memory interfaces completely and integrate all 8 or so GB)?
     
  8. McHuj

    Veteran Regular Subscriber

    Joined:
    Jul 1, 2005
    Messages:
    1,381
    Likes Received:
    497
    Location:
    Texas
    Should be by 2015. Most of the fabs have been developing 16nm/14nm in parallel to 20nm as they need to close the gap with Intel. (As much as they can at least). Hopefully they won't split and 20nm might actually be a pretty short lived node.
     
  9. McHuj

    Veteran Regular Subscriber

    Joined:
    Jul 1, 2005
    Messages:
    1,381
    Likes Received:
    497
    Location:
    Texas
    Yeah, I believe so. To achieve 1 TB/sec, they would need an extremely wide bus like 1024-bits at 8 Gbps IO. I think that would be very cost prohibitive.
     
  10. Homeles

    Newcomer

    Joined:
    May 25, 2012
    Messages:
    234
    Likes Received:
    0
    GDDR6 would be out by then, so the scenario isn't quite as grim.
     
  11. iMacmatician

    Regular

    Joined:
    Jul 24, 2010
    Messages:
    759
    Likes Received:
    198
    On the GPU roadmap slide, is the y-axis DP GFLOPS/W? Seems like it at least from Fermi and Kepler's placements.
     
  12. UniversalTruth

    Veteran

    Joined:
    Sep 5, 2010
    Messages:
    1,747
    Likes Received:
    22
    Yup, I remember discussing this here. If so, I guess late 2015/ early 2016 for Volta.
     
  13. lanek

    Veteran

    Joined:
    Mar 7, 2012
    Messages:
    2,469
    Likes Received:
    315
    Location:
    Switzerland
    It is .. But the placement is large .. somewhere between 8-16 and 16 -32
    [​IMG]

    [​IMG]
     
  14. Ninjaprime

    Regular

    Joined:
    Jun 8, 2008
    Messages:
    337
    Likes Received:
    1
    Looks like all the memory is stacked, or at least they are showing 24 RAM chips, which is about right.
     
  15. DavidGraham

    Veteran

    Joined:
    Dec 22, 2009
    Messages:
    2,613
    Likes Received:
    2,222
    So what advantage does Stacked memory hold over standard memory ? less latency? faster access times? or just plain old increase in frequency ?
     
  16. gkar1

    Regular

    Joined:
    Jul 20, 2002
    Messages:
    614
    Likes Received:
    7
  17. mczak

    Veteran

    Joined:
    Oct 24, 2002
    Messages:
    3,004
    Likes Received:
    109
    What happened to Einstein, the designated successor of Maxwell? Is Volta an entirely different project, with a different architecture?
     
  18. ninelven

    Veteran

    Joined:
    Dec 27, 2002
    Messages:
    1,693
    Likes Received:
    107
    I think Einstein was/is planned for 10nm, so Volta would be the chip that falls between Maxwell and Einstein. I imagine it will just be a refinement of Maxwell with the largest benefit being the stacked memory.
     
  19. DSC

    DSC
    Banned

    Joined:
    Jul 12, 2003
    Messages:
    689
    Likes Received:
    3
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...