And here I thought I had a stable overclock...

Discussion in 'Folding@Home - Join Team Guru3D !' started by J_J_B, Oct 28, 2010.

  1. J_J_B

    J_J_B Member Guru

    Messages:
    159
    Likes Received:
    0
    GPU:
    EVGA: 2xSLI 580HC, 480HC
    First post in ages... never stopped reading Guru3D's articles but only just returned to folding after slowly accumulating parts for my first water-cooled system since January, which was assembled and powered-up without incident about a month ago. Sold my two GTX-295 cards many months ago and now just need a second GTX-480 to get the system to a point where gaming frame rates exceed what I had. Might pick up a 3rd in mid to late 2011 but suspect I'll need a second PSU if I do that. System is currently pulling 675 watts while folding on GPU3 and SMP2 clients. I designed in plenty of radiators for the GPU cooling loop but didn't factor in my desire to overclock these new GPUs!

    I thought I had a stable 4.26-GHz CPU overclock on this system until I started folding again. That's when I found that my long-term stability wasn't there... system would blue-screen after 8 to 36 hours of folding with 4 of the standard x86 tray clients running. I kept increasing the VCore until I reached 1.45V with vdroop disabled. That helped, but didn't eliminate the problem. I decided to pull back to 4.12-GHz rather than risk 1.5V and the system is running like a charm now. I think the 4.1-GHz "cap" can be chaulked up to the C0 stepping on this original generation Core i7 965 chip.

    I currently have the SMP2 beta client running at 10500 PPD. The first two work units completed with performance fractions of "0.907765" and "0.934677". I think I read here that there is a bonus if 10 consecutive clients are returned with performance fractions exceeding "0.8". Is that correct?

    On the GPU side, I'm running the GPU3 beta client. This single GTX-480 was folding at close to 19000 PPD with it overclocked to 900/1800/2150. Initially I had the memory clocked at 2200. Once again, I thought I had a stable overclock but quickly noticed that an error was occuring and the work units were being returned within a minute. I slid the memory overclock down to 2150 and things have been fine ever since. I say that it "was" at 19000 PPD because I now have the SMP2 beta client running. Even with the "-smp 7" option, I see the GPU usage jittering. For instance, as I write this post I can see it jitting between 100% and 25%. I should also note that I have an encrypted Virtual Machine running where I do all my work office work, so that also affects things too... though to a lesser degree than my working/typing in an app running directly on the host. I initially had the SMP2 client running on all 8 threads but noticed that the GPU usage was hovering at only 50%. By using the "-smp 7" option I got the usage closer to 100% again and the rate on the GPU is up to around 17500 PPD.

    So... I'm happy to be folding again and just as happy to have solved some stability issues thanks to the F@H clients!
     
    Last edited: Oct 28, 2010
  2. PantherX

    PantherX Folding@Home Team Leader Staff Member

    Messages:
    1,380
    Likes Received:
    0
    GPU:
    Nvidia GTX 1080 Ti
    Regarding the Bonus Points, it is like this:
    Use a passkey with the SMP2 BETA Client v6.30
    Fold 10 FahCore_a3 WUs before the preferred deadline
    Maintain a >=80% return rate of the WUs assigned to you (it isn't performance fraction)

    BTW, folding is stressful on the system so one may have to tweak their OC to get it F@H Stable. If you have used Prime95 as a stability test, then it is useless since folding is more stressful. Nonetheless, it is nice that your system is stable.
     
  3. iancook221188

    iancook221188 Ancient Guru

    Messages:
    1,725
    Likes Received:
    11
    GPU:
    GTX 670 SLI / GTX 460 SLI
    nice core overclock on 480 (do you have the temp you are running at), overclocking the memory will have no affect on folding so it up to you if you need/want to run it that speed

    have you tried to vdroop on i have it on, on my 970 and im current at 4Ghz 1.206v it mite be like you said though the CO stepping wall
     
  4. J_J_B

    J_J_B Member Guru

    Messages:
    159
    Likes Received:
    0
    GPU:
    EVGA: 2xSLI 580HC, 480HC
    Thanks Panther. I am using a passcode so all should be well. I did indeed use Prime95 as my CPU stresser during testing. I guess I should have been using the Intel Burn Test utility, which I believe I read was even more stressfull than F@H.

    iancook:

    I havn't actually tried enabling the vdroop since what I had read suggested keeping it disabled, in which case the voltage settings in the BIOS become "minimums".

    Temperature on the GPU is maxing out at 45 degrees C while folding. When running Kombuster it was peaking around 51 degrees C. This is at an office temperature of 20 degrees C. I'm using the EVGA e-leet utility to set the GPU voltage to 1.175V and the PWM to 416 kHz to acquire the 900/1800/2150 overclock. I think I can safely go a little higher now that the memory has been pulled back a bit. Just set it to 925/1850 but I'll want to monitor the F@H client for a while and run Kombuster later to ensure that it really is stable. I'll try 950 after that because I really am surpised by how much the PPD rate just jumped with that little bit of change... up to 18200 from 17500 on the current WU.

    I only have one GTX-480 in this chassis right now and tons of cooling for it: two RX120 rads and one RX240 (XSPC) for the graphics loop alone. CPU loop has a single XSPC RX360 rad.
     
    Last edited: Oct 28, 2010

  5. J_J_B

    J_J_B Member Guru

    Messages:
    159
    Likes Received:
    0
    GPU:
    EVGA: 2xSLI 580HC, 480HC
    Tried overclocking the GTX-480 to 940/1880/2150 but it didn't like it (GPU3 client errored on 3 work units while I drove home last evening). Dropped the overclock back to 925/1850/2150 and things seem stable there. I'll still need to run Kombuster for a few hours to prove to myself that it really is stable but the F@H client seems happy for now.

    At the moment the folding rate on this system with P10632 (GPU3) and a P6071 (SMP2) work units folding is 33954.6 PPD.

    I'm wondering...

    When I enabled the SMP2 beta client I had to trim the number of threads in use by the client from 8 to 7 in order to get the GTX-480 usage up from 50% to 100% (note: I configured the SMP2 client to set it's priority to "low" instead of "idle"). When I eventually install a second GTX-480, will that same thread that I freed up be able to accomodate the usage of the second GPU at 100% or do you think I'll need to choke back the SMP2 client again (to 6 threads)?

    I don't think I'll be able to afford the second GTX-480 hydro copper until February, unless I sell some old network equipment. Just thinking ahead here...
     
    Last edited: Oct 29, 2010
  6. PantherX

    PantherX Folding@Home Team Leader Staff Member

    Messages:
    1,380
    Likes Received:
    0
    GPU:
    Nvidia GTX 1080 Ti
    The best answer is to test it yourself. However, my guess is that 2 GTX 480 will use that free thread. Do note that you might loose a few PPD but generally, the rule is that PPD Gain from GPU > PPD loss in SMP2. In certain cases, you have to tweak the system, i.e. -smp 6 VS -smp 7 to see which one gives you the lowest TPF on your GPUs and SMP2. Personally, I wouldn't be that much bothered since the upgrade is in February so things may change on the F@H side.
     
  7. iancook221188

    iancook221188 Ancient Guru

    Messages:
    1,725
    Likes Received:
    11
    GPU:
    GTX 670 SLI / GTX 460 SLI
    i think im going to need 4.2 - 4.4 to get 99-100% on both gpu3 on one thread, clinet 11265 porject i.e 912 925 pointer the 611 pointers use a lot less cpu power so ther efine on one thread. at 4Ghz im see 75-85-97% gpu usage and on the other im see 98-99% gpu usage
     
  8. J_J_B

    J_J_B Member Guru

    Messages:
    159
    Likes Received:
    0
    GPU:
    EVGA: 2xSLI 580HC, 480HC
    Thanks for the info (both of you!). I can't see me upgrading to a 980X (better overclock potential, plus the 2 extra cores) until the prices come way down. Back when I purchased the 965 I wasn't married... and didn't have a kid on the way!

    ... yet here I am talking like that while saving up for a second GTX-480 hydro copper, possibly two.
     
  9. J_J_B

    J_J_B Member Guru

    Messages:
    159
    Likes Received:
    0
    GPU:
    EVGA: 2xSLI 580HC, 480HC
    FYI: I had a single GPU3 work unit error out on me the other night so I pulled the GTX-480 HC back to what I know to be safe on my card: 900/1800/2150 at a voltage of 1.175, PWM frequency of 416 (using EVGA e-leet utility). Weeks ago I had experimented with higher voltages but found that increasing the voltage beyond 1.175 did not provide any extra stability.

    Earlier in this thread I mentioned that my GPU temp while running Kombuster topped out at 51 degrees C. That was incorrect... it tops out at 53 degrees C (3-hr session).

    Upgraded to the latest version of the EVGA e-leet utility yesterday and noticed that the maximum GPU voltage was 1.135 volts. I quickly uninstalled it and re-installed the beta version from early October so I could regain my 1.175 volts. It's a nuisance having to manually start e-leet to set the voltage before using the EVGA Precision utility to overclock. I'd use Afterburner but it also has the 1.135 volt limit, and my understanding is that this limit may actually be being pulled from my card's BIOS rather than something that is hardcoded into these 3rd-party utilities. Is that correct?
     
  10. PantherX

    PantherX Folding@Home Team Leader Staff Member

    Messages:
    1,380
    Likes Received:
    0
    GPU:
    Nvidia GTX 1080 Ti

Share This Page