Stanford's Mike Houston on F@H GPU client

Discussion in 'GPGPU Technology & Programming' started by Geo, Sep 29, 2006.

Page 4 of 6
  1. Jawed

    Jawed
    Legend

    Joined:
    Oct 2, 2004
    Messages:
    11,716
    Likes Received:
    2,137
    Location:
    London
    No. The program spends most of its time just looping to perform calculations.

    There's prolly a minor effect relating to bandwidth, but that'll only come out after a few days of testing...

    Jawed
     
    #61 Jawed, Oct 2, 2006
  2. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    215
    Location:
    Uffda-land
    Yay, and I already had an EUE on my first WU. Woot, woot, beta testers do it for teh love of the game. :wink:
     
    #62 Geo, Oct 2, 2006
  3. ElMoIsEviL

    ElMoIsEviL
    Newcomer

    Joined:
    Nov 3, 2003
    Messages:
    21
    Likes Received:
    0
    Location:
    Ottawa, Canada
    Mine's doing -12C on load with Folding@Home running.. still much better then the -5C I hit when gaming.

    (ThermoElectrically cooled VPU's).

    Now if they could only support Crossfire I'd be happy.. :)
     
    #63 ElMoIsEviL, Oct 2, 2006
  4. Jawed

    Jawed
    Legend

    Joined:
    Oct 2, 2004
    Messages:
    11,716
    Likes Received:
    2,137
    Location:
    London
    Yep already seen one guy say he's sending back his new X1900XT (unopened) because the PPD are so low compared with his C2D (well over twice as many PPD per core).

    Jawed
     
    #64 Jawed, Oct 2, 2006
  5. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    215
    Location:
    Uffda-land
    Point at this conversation please?
     
    #65 Geo, Oct 3, 2006
  6. Jawed

    Jawed
    Legend

    Joined:
    Oct 2, 2004
    Messages:
    11,716
    Likes Received:
    2,137
    Location:
    London
    #66 Jawed, Oct 3, 2006
  7. willardjuice

    willardjuice super willyjuice
    Moderator Veteran Alpha

    Joined:
    May 14, 2005
    Messages:
    1,386
    Likes Received:
    299
    Location:
    NY
    Someone with power ask Terry to release the beta in x64 flavor. :cry:
     
    #67 willardjuice, Oct 3, 2006
  8. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    215
    Location:
    Uffda-land
    Well, I've had two EUE in a row (Early Unit End), so I guess I should try to find that forum Vijay was referring to.
     
    #68 Geo, Oct 3, 2006
  9. Bohdy

    Bohdy
    Regular

    Joined:
    Jun 9, 2003
    Messages:
    731
    Likes Received:
    4
    Lol, that is a problem. People will be avoiding GPU folding because they can get more points on their CPU's, even though the GPU is doing far more work.

    The folding guys really need to up the reward for GPU folding, for those only concerned with points. I don't think making it "fair" for CPU folders is necessary, as the goal of the project is maximum folding performance. People will buy X1900's to catch up.
     
    #69 Bohdy, Oct 3, 2006
  10. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    215
    Location:
    Uffda-land
    Interesting. I gather PPD is Points Per Day? I'm impressed that he's getting 1,500/day with his current CPU, as it seems to me I do about 1,500/week with an OC'ed X2 that runs 24/7. My overall is higher than that, because I've got two other slower part-time cpus that chip in here and there.

    Points are great, and I hope they get it worked out, but at the end of the day I'm trying to cure disease, and that's going to be the touchstone for what client I run.
     
    #70 Geo, Oct 3, 2006
  11. Jawed

    Jawed
    Legend

    Joined:
    Oct 2, 2004
    Messages:
    11,716
    Likes Received:
    2,137
    Location:
    London
    The forum I linked earlier is the forum you want!

    It seems that Pande is recommending people use Cat 6.5. Read the threads on EUE for more...

    Jawed
     
    #71 Jawed, Oct 3, 2006
  12. Jawed

    Jawed
    Legend

    Joined:
    Oct 2, 2004
    Messages:
    11,716
    Likes Received:
    2,137
    Location:
    London
    It seems that a Kentsfield (4 core) is scoring about 35x what the base P4 2.8GHz (110PPD) can score - i.e. around 4000PPD. And a C2D is scoring about 20x or more (2000-2500PPD).

    So if an X1900XTX/single-core-CPU (since both together produce a GPU-WU) is about 30x faster than the base P4, but can do a smaller range of WUs, then it's prolly worth less than 30x points. But ~1/8th?...

    It seems to me that GPGPU isn't quite the monstrous revolution in F@H computing that it's been cracked up to be. Core 2 has turned the tables...

    CTM to the rescue?

    Jawed
     
    #72 Jawed, Oct 3, 2006
  13. Bohdy

    Bohdy
    Regular

    Joined:
    Jun 9, 2003
    Messages:
    731
    Likes Received:
    4
    Whoah Jawed, hang on. First, how do you know that they meant the GPU core was 20-30x faster compared to the benchmark machine?
     
    #73 Bohdy, Oct 3, 2006
    Last edited by a moderator: Oct 3, 2006
  14. RacingPHT

    RacingPHT
    Newcomer

    Joined:
    May 27, 2006
    Messages:
    90
    Likes Received:
    8
    Location:
    Shanghai
    interesting... I'm just looking forward to CTM.

    mhouston, can you tell us how scatter works? :)
    what if the GPU write to the same location? and what is the performance of scattering?
     
    #74 RacingPHT, Oct 3, 2006
  15. Jawed

    Jawed
    Legend

    Joined:
    Oct 2, 2004
    Messages:
    11,716
    Likes Received:
    2,137
    Location:
    London
    Yeah.

    Jawed
     
    #75 Jawed, Oct 3, 2006
  16. Bohdy

    Bohdy
    Regular

    Joined:
    Jun 9, 2003
    Messages:
    731
    Likes Received:
    4
    "Yeah" is a very ambivalent answer to the question :razz:
     
    #76 Bohdy, Oct 3, 2006
  17. Geo

    Geo Mostly Harmless
    Legend

    Joined:
    Apr 22, 2002
    Messages:
    9,116
    Likes Received:
    215
    Location:
    Uffda-land
    Well, I abandoned 6.10 on Stanford's advice, and 6.5 is happily puttering along. One thing I noticed is that any change to 3D settings in CCC while the client is running will cause it to blow up, tho it will resume from the latest checkpoint when you start it again.
     
    #77 Geo, Oct 3, 2006
  18. mhouston

    mhouston A little of this and that
    Regular

    Joined:
    Oct 7, 2005
    Messages:
    344
    Likes Received:
    38
    Location:
    Cupertino
    Yep, anytime the context is taken away from us we have to restart at the last checkpoint. Any mode changes seem to cause this, although some screen savers which don't lock the screen, seem to behave just fine. Most full-screen games will dump us out.
     
    #78 mhouston, Oct 3, 2006
  19. Bohdy

    Bohdy
    Regular

    Joined:
    Jun 9, 2003
    Messages:
    731
    Likes Received:
    4
    #79 Bohdy, Oct 3, 2006
  20. BRiT

    BRiT (>• •)>⌐■-■ (⌐■-■)
    Moderator Legend Alpha

    Joined:
    Feb 7, 2002
    Messages:
    20,516
    Likes Received:
    24,424
    It's a true shame that the PPD makes GPU folding not worth it. :( Strictly from a PPD perspective, with a modern day CPU, one can achieve significantly more when the core is dedicated to cpu-folding than assiting in GPU-folding.
     
    #80 BRiT, Oct 4, 2006
(You must log in or sign up to post here.)
Page 4 of 6
Loading...
Similar Threads - Stanford's Mike Houston
  1. Rys

    Beyond Programmable Shading as CS448s at Stanford this year

    Rys, Apr 2, 2010, in forum: GPGPU Technology & Programming
    Replies:
    13
    Views:
    10,618
    mhouston
    May 26, 2010
  2. B3D News

    Stanford previews new F@H GUI and GPU client at SuperComputing 2007

    B3D News, Nov 12, 2007, in forum: GPGPU Technology & Programming
    Replies:
    27
    Views:
    15,225
    Arnold Beckenbauer
    Apr 10, 2008
  3. one

    SCE Joins Stanford's Folding@home Program (B3D ID=32377)

    one, Mar 15, 2007, in forum: Console Industry
    Replies:
    409
    Views:
    47,932
    ShaidarHaran
    May 19, 2008
  4. Geo

    Stanford adds X1800 and X1600 support to FAH

    Geo, Dec 4, 2006, in forum: Beyond3D News
    Replies:
    4
    Views:
    2,624
    digitalwanderer
    Dec 6, 2006
  5. Geo

    Stanford and ATI release F@H clients and drivers

    Geo, Oct 2, 2006, in forum: Beyond3D News
    Replies:
    1
    Views:
    1,860
    turingo
    Oct 5, 2006
  6. XavierS

    Peter Hofstee talk at Stanford

    XavierS, Feb 26, 2006, in forum: PC Hardware, Software and Displays
    Replies:
    1
    Views:
    1,529
    Pete
    Feb 27, 2006

Share This Page

Toggle Width
Style
Beyond3D

  • 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.

  • Quick Navigation

Loading...