Getting back into folding...

Discussion in 'Distributed Computing' started by Athlex, Feb 28, 2017.

  1. Athlex

    Athlex Golden Member

    Joined:
    Jun 17, 2000
    Messages:
    1,258
    Likes Received:
    2
    Been a few years but with this cool weather, I thought I'd fire up F@H again and do something useful while keeping the room warm. :)

    My system is relatively old (Phenom II X6 1090T 3.2GHz @ 3.8GHz) and from what I'm reading, CPU folding isn't really worth much at all compared with GPUs. Is it worth running the CPU at all on a couple cores or is it just a detriment to the GPUs? It's funny looking through team stats since I'm ranked 55th in points and 4th in WU from the horde of [mostly P4] machines I used to fold with...

    GPU-wise, I have a pair of identical Sapphire R9-280 boards in an AMD 970 (x16 and x4) crossfire setup. Looks like PCIe bandwidth is definitely a bottleneck since I'm getting ~18,000 PPD from one board, and ~33500 PPD from the other.

    Couple questions for the pros here-
    • Does enabling/disabling Crossfire matter? Output seems to float between 40,000 and 44,000 PPD with or without crossfile with both GPUs running and the CPU paused, so I'm guessing "no?"
    • I'm reading good things about Windows 10 versus Win 7 for folding. I'm currently on Win 8.1, any idea if I'd see an improvement if I were to upgrade?
    • Any other suggestions for getting the most out of this setup? Efficiency-wise, it seems like running the faster GPU is ideal, but also wondering what other things I'm not considering like the value of WUs from different projects.
    • Edit: Any benefit to disabling programs like f.lux while Folding?
    Thanks for any tips!
     
    #1 Athlex, Feb 28, 2017
    Last edited: Feb 28, 2017
    Ken g6 and Smoke like this.
  2. Loading...

    Similar Threads - back folding Forum Date
    Back again and help needed.*Fixed* Distributed Computing Oct 1, 2015
    Going back to Folding and after a few tweaks... Distributed Computing Nov 16, 2013
    Back Folding for team 198 Distributed Computing Apr 13, 2013
    Folding@Home Stats for 2002-Nov-28 to 2002-Dec-02: Back to 17th we go! Distributed Computing Dec 2, 2002
    Folding@Home Stats for 2002-Nov-27: We fell back to 18th today Distributed Computing Nov 27, 2002

  3. StefanR5R

    StefanR5R Member

    Joined:
    Dec 10, 2016
    Messages:
    181
    Likes Received:
    79
    Regarding PCIe bandwidth: GPU-Z and HWiNFO64 are showing a "bus interface load" percentage (with Nvidia 10-series GPUs at least) which I presume is utilization of the PCIe bus. With my comparably fast cards (GTX 1070, 1080), this bus interface load during typical F@H WUs is shown at about 33 % for PCIe 3.0 x16 and somewhat less than 50 % for PCIe 3.0 x8. So, extrapolating from PCIe 3.0 to PCIe 2.0, x16 and x8 should still be fine, but PCIe 2.0 x4 is likely to be a bottleneck at least for similar Nvidia cards.

    Also, from the thread "Getting the most PPD out of your hardware for F@H":

    Your numbers indicate that the same is true with AMD R9 cards.

    (The "Getting the most PPD..." thread has a lot more info of this kind.)
     
    Smoke likes this.
  4. Markfw

    Markfw CPU Moderator VC&G Moderator Elite Member
    Super Moderator

    Joined:
    May 16, 2002
    Messages:
    14,196
    Likes Received:
    400
    I am slightly confused on ppd. My R9 290 was doing about 220,000 ppd for ONE card. Is the 280 that much slower ?

    I don't know about crossfire, but SLI is supposed to work, but does create problems.
     
  5. Athlex

    Athlex Golden Member

    Joined:
    Jun 17, 2000
    Messages:
    1,258
    Likes Received:
    2
    Good grief, that's a huge difference. With both GPUs and CPU running, I'm consistently in the 55,000PPD range. From what I'm seeing, the R9 280 is basically 2012-era silicon since that drive the 79x0, 8950, and R9 280, so I guess it's just that old.

    I've also been experimenting with Crossfire enabled and disabled over the last day or two and I'm not seeing any significant difference, so I guess it's not as temperamental as SLI.