[computerworld] Nvidia, AMD release graphics processors for supercomputing

Discussion in 'Video Cards and Graphics' started by Final8ty, Nov 10, 2012.

  1. Final8ty

    Final8ty Golden Member

    Joined:
    Jun 13, 2007
    Messages:
    1,165
    Likes Received:
    3
    Nvidia's K20 and K20X, and AMD's FirePro SM10000 GPUs deliver more than a teraflop of double-precision performance

    http://www.computerworld.com.au/art...d_release_graphics_processors_supercomputing/
     
  2. Anarchist420

    Anarchist420 Diamond Member

    Joined:
    Feb 13, 2010
    Messages:
    8,649
    Likes Received:
    0
    It would be nice if nv actually cared about DP for consumer parts.

    Good thread BTW:)
     
  3. Gryz

    Gryz Golden Member

    Joined:
    Aug 28, 2010
    Messages:
    1,141
    Likes Received:
    4
  4. Keysplayr

    Keysplayr Elite Member

    Joined:
    Jan 16, 2003
    Messages:
    21,120
    Likes Received:
    22
    What would that net you?
     
  5. Plimogz

    Plimogz Senior member

    Joined:
    Oct 3, 2009
    Messages:
    678
    Likes Received:
    0
    K20X? so I guess ORNL received the best chips, those with 14/15 SMX functional, for Titan while there'll be a K20 with 13/15 and a memory controller disabled.

    It'll be interesting to see if and when nV revise and re-spin big Kepler to achieve what they did going from 480 to 580, in order to get some fully functional chips out to market.

    But that Sea Islands based SM10000 sounds like fun. And it bodes well for the 89xx Radeons being here well before next summer, too.
     
  6. Alusan

    Alusan Member

    Joined:
    Mar 5, 2010
    Messages:
    33
    Likes Received:
    0
    I think it's more likely that it's that dual-Tahiti board with downclocked GPUs they showed a few months back. Wouldn't we have heard some more rumblings about Sea Islands if they already had a FirePro product ready?
     
  7. DiogoDX

    DiogoDX Senior member

    Joined:
    Oct 11, 2012
    Messages:
    572
    Likes Received:
    0
    I agree. SM10000 is a Seamicro Server not an evolution of the S9000.

    http://www.amd.com/us/press-releases/Pages/amd-based-hp-servers-2012oct9.aspx
     
  8. Homeles

    Homeles Platinum Member

    Joined:
    Dec 9, 2011
    Messages:
    2,585
    Likes Received:
    0
    It's a fully enabled Tahiti chip, running at lower clocks than the 7970. It should be S10000, not SM10000.
     
  9. Alusan

    Alusan Member

    Joined:
    Mar 5, 2010
    Messages:
    33
    Likes Received:
    0
    Do you mean two Tahiti chips? The article mentions 1.48TFLOPs DP performance.
     
  10. Plimogz

    Plimogz Senior member

    Joined:
    Oct 3, 2009
    Messages:
    678
    Likes Received:
    0
  11. Homeles

    Homeles Platinum Member

    Joined:
    Dec 9, 2011
    Messages:
    2,585
    Likes Received:
    0
    No. GCN has FP64 up to ½ of its FP32 rate. Consumer cards max out at ¼.
     
  12. ViRGE

    ViRGE Elite Member, Moderator Emeritus

    Joined:
    Oct 9, 1999
    Messages:
    31,108
    Likes Received:
    16
    When AMD calls FP64 performance configurable, they don't mean in the "flip a software switch sense". It's configurable in that it's easy to modify a design to have higher or lower FP64 performance based on your design goals (higher performance requiring larger hardware blocks). Pitcairn and Cape Verde are 1/16, Tahiti is 1/4. There is not a Southern Islands chip with 1/2. That's merely an architectural option for AMD if they want to go that way on future chips.

    So based on this article, it has to be a dual-GPU card. 1.48 TFLOPS is not possible on a single Tahiti GPU.
     
  13. 3DVagabond

    3DVagabond Lifer

    Joined:
    Aug 10, 2009
    Messages:
    11,730
    Likes Received:
    2
    TechPowerUp has an article on the S10000 server graphics card.

    [​IMG]

    Dual GPU card.
     
  14. Lonbjerg

    Lonbjerg Diamond Member

    Joined:
    Dec 6, 2009
    Messages:
    4,426
    Likes Received:
    0

    Nice PR spin form your link:
    I guuess CUDA dosn't get recognized by AMD...go figure ;)



    No fanboy-fueled trolling please.

    In the future, I hope you won't feel obligated to troll everything AMD-related.

    Moderator jvroig
     
    #14 Lonbjerg, Nov 12, 2012
    Last edited by a moderator: Nov 12, 2012
  15. Keysplayr

    Keysplayr Elite Member

    Joined:
    Jan 16, 2003
    Messages:
    21,120
    Likes Received:
    22
    Ok, ya lost me here. What does the AMD GPU(s) 3dVagabond listed have to do with CUDA? Should it? No idea where you were coming from with this. Blame it on my first cup of coffee if you wish.

    Can you clear this up for the bleary eyed?
     
  16. Jaydip

    Jaydip Diamond Member

    Joined:
    Mar 29, 2010
    Messages:
    3,615
    Likes Received:
    3
  17. sontin

    sontin Diamond Member

    Joined:
    Sep 12, 2011
    Messages:
    3,016
    Likes Received:
    0
  18. Jaydip

    Jaydip Diamond Member

    Joined:
    Mar 29, 2010
    Messages:
    3,615
    Likes Received:
    3
    S10000 has better DP and SP performance but at the expense of TDP.
     
    #18 Jaydip, Nov 12, 2012
    Last edited: Nov 12, 2012
  19. sontin

    sontin Diamond Member

    Joined:
    Sep 12, 2011
    Messages:
    3,016
    Likes Received:
    0
    S10000 has only 13% more DP performance and needs ~66% more power. That's not impressive.
     
  20. Jaydip

    Jaydip Diamond Member

    Joined:
    Mar 29, 2010
    Messages:
    3,615
    Likes Received:
    3
    AMD needed to create a Tesla equivalent can't blame them for trying.
     
  21. Silverforce11

    Joined:
    Feb 19, 2009
    Messages:
    10,458
    Likes Received:
    1
    Doesnt matter, AMD wont even make a blip in HPC, ppl are already used to CUDA.

    At least Intel has the muscle and $$ to hammer away at its x86 for their HPC attempt. AMD has nothing of the sort to dent away at NV's dominance.
     
  22. Jaydip

    Jaydip Diamond Member

    Joined:
    Mar 29, 2010
    Messages:
    3,615
    Likes Received:
    3
    Gotta disagree.You will get nowhere unless you try. Amd needs to offer more than hardware, they need to offer a complete package.The potential is there,they got to take advantage of it.
     
  23. Silverforce11

    Joined:
    Feb 19, 2009
    Messages:
    10,458
    Likes Received:
    1
    They are bleeding $$ as is, wasting resources fighting NV in the HPC sector, they are doomed to fail.
     
  24. 3DVagabond

    3DVagabond Lifer

    Joined:
    Aug 10, 2009
    Messages:
    11,730
    Likes Received:
    2
    CUDA is proprietary. nVidia has to support it. OpenCL is open source. In theory anyway, it doesn't need vendor support.
     
  25. Lonbjerg

    Lonbjerg Diamond Member

    Joined:
    Dec 6, 2009
    Messages:
    4,426
    Likes Received:
    0
    So just like Direct3D vs OpenGL....that went well for the open source...oh wait.

    It's more complicated that just tossing hardware out there....hence why CUDA is used in schools ^^
     
Loading...