GCN Radeons for crunching

Discussion in 'Distributed Computing' started by dkm777, Apr 5, 2012.

  1. dkm777

    dkm777 Senior member

    Joined:
    Nov 21, 2010
    Messages:
    528
    Likes Received:
    0
    Hi guys,

    Is it possible to use GCN based cards for DC yet? I'm currently considering whether to get myself a HD7870 and can't find anything related to DC performance on GCN cards. Before launch AMD touted them as vastly superior in computing to old VLIW Radeons. Is there a project that added support recently?
     
  2. Loading...

    Similar Threads - Radeons crunching Forum Date
    Radeon 4770 Distributed Computing Feb 19, 2017
    Radeon R9 200 series on Vista Distributed Computing Dec 19, 2016
    Old AMD Radeon R9-280X prices Distributed Computing Jan 2, 2016
    Radeon Fury X Distributed Computing Jun 16, 2015
    GPU crunching with Radeon 3850? Distributed Computing Nov 16, 2007

  3. Sunny129

    Sunny129 Diamond Member

    Joined:
    Nov 14, 2000
    Messages:
    4,821
    Likes Received:
    0
    i know Milkyway@Home runs on 79xx/78xx/77xx series GPUs. but keep in mind that MW@H makes uses of double precision (FP64) calculations. as such, only a 79xx series card is going to be faster than the older VLIW4 and VLIW5 architectures b/c its FP64 performance isn't crippled like it is on a 78xx/77xx series GPU. like the VLIW4 architecture, the FP64 performance of the 79xx series is 1/4 of its single precision (FP32) performance. the FP64 performance of the 78xx and 77xx series is limited to 1/16 of its single precision (FP32) performance. for comparison, a 77xx/78xx series GPU would have to be ~3.2 times more powerful than an HD 5870 for its FP64 performance to match that of the HD 5870's VLIW5 architecture (who's FP64 performance is only 1/5 of its FP32 performance). so if you plan on participating in a project that requires double precision (MW@H is the only one i know of right now), you can use a 78xx series GPU, but it'll be far less efficient than an HD 58xx/69xx series GPU. to get a raw comparison of double precision floating point compute power, take a look at these ATI GPU comparison charts.

    i think Collatz Conjecture is now running on 7xxx series GPUs, and perhaps PrimeGrid and GPUGrid, but you'll have to visit their websites to confirm it.
     
  4. dkm777

    dkm777 Senior member

    Joined:
    Nov 21, 2010
    Messages:
    528
    Likes Received:
    0
    Oh well, got a GTX570 instead, because it was cheaper.
     
  5. Sunny129

    Sunny129 Diamond Member

    Joined:
    Nov 14, 2000
    Messages:
    4,821
    Likes Received:
    0
    that's great! a GTX 570 is a quite powerful cruncher on a number of projects...
     
  6. blckgrffn

    blckgrffn Diamond Member

    Joined:
    May 1, 2003
    Messages:
    6,647
    Likes Received:
    2
    Indeed! :thumbsup:

    Very nice :) We'll be looking for you in the upcoming PrimeGrid challenges that heavily favor GPUs!
     
  7. dkm777

    dkm777 Senior member

    Joined:
    Nov 21, 2010
    Messages:
    528
    Likes Received:
    0
    I gave it a workout with Folding before leaving for work. Let me tell you those 7000 series projects are big bastards. FAHMon says it will take over 12 hours for my GPU to finish. Prime Grid is on BOINC, right?
     
  8. Sunny129

    Sunny129 Diamond Member

    Joined:
    Nov 14, 2000
    Messages:
    4,821
    Likes Received:
    0
    yes, PrimeGrid is a BOINC project.