Message boards :
Number crunching :
Getting the most performance (with overclocking)
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 16 Jul 12 Posts: 98 Credit: 386,043,752 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
Are there any tips or tricks one can do to squeeze the last bit of performance out of the GPU for crunching? I already allocated 2.5 CPU cores for my 2 GPU's and overclocked the CPU to ensure no bottlenecks, even though the CPU is fairly modern (i7-3820). In terms of overclocking, I know that overclocking can cause tasks to fail, I have experienced this myself, however, I was wondering if it was safe to do a really small overclock, as in +25 mhz to the GPU clock. Would this cause tasks to fail? I figure that even though it's a small overclock, if it's stable, the tiny performance gain given by it will add up over the days/months/etc of crunching. To say it simply: How much of an overclock can be pulled off that wouldn't fail tasks? Even a 1 mhz overclock would do something. |
microchipSend message Joined: 4 Sep 11 Posts: 110 Credit: 326,102,587 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
There is no general rule to overclocking. It depends on various factors, such as quality of MB, CPU, RAM, etc. The best rule is to try and see which one is stable for your specific PC configuration. Team Belgium |
|
Send message Joined: 16 Jul 12 Posts: 98 Credit: 386,043,752 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
True, but a stable GPU overclock would still fail tasks for me on GPUGRID, but perform fine on other tasks. I want to know how far one could overclock while still successfully completing tasks. If it helps, I could post my system specs. |
microchipSend message Joined: 4 Sep 11 Posts: 110 Credit: 326,102,587 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
With GPUGRID it's very tricky to know in advance which overclock will be stable for it. GPUGRID is very sensitive to overclocking and temperatures. For example, my GPU a while ago was running at close to 80°C and it repeatedly failed on GPUGRID while it had no problems on Milkyway and Einstein. Increasing the fan speed so it lowered the temps to 65°C was enough to make it stable for GPUGRID Sorry if this doesn't answer your question but I honestly can't guess which OC will be stable for GPUGRID tasks. Maybe someone else can chime in... Team Belgium |
|
Send message Joined: 16 Jul 12 Posts: 98 Credit: 386,043,752 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
Thanks, that answer did help. I read in another thread that the actual WU can have some effect on the stability. I have two cards at around 65-70C, so I'm going to put a 25 Mhz overclock on each GPU clock and see where that gets me. Thanks again. |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
For GPU-Grid pretty much the only OC that matters would be the chip clock of your GTX670s. You'll probably reach ~1.20 GHz at 1.175 V (maximum Turbo bin, maximum available via EVGA Precision) stable at GPU-Grid. Mine (GTX660Ti) can do 1.23 GHz here, POEM at 1.33 GHz.. but will fail GPU-Grid at 1.25 GHz. SK's is doing ~20 MHz less and yours will be slightly different as well. MrS Scanning for our furry friends since Jan 2002 |
|
Send message Joined: 16 Jul 12 Posts: 98 Credit: 386,043,752 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]()
|
Alright, thanks for the answer, I'll play around and see what works. |
|
Send message Joined: 17 Aug 08 Posts: 2705 Credit: 1,311,122,549 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Just make sure you'Re not running one of the current NOELIA WUs whilöe experimenting, these are often failing even on otherwise good hosts. MrS Scanning for our furry friends since Jan 2002 |
©2025 Universitat Pompeu Fabra