Message boards :
Number crunching :
New long runs NOELIAs in my 750Ti
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 25 Mar 12 Posts: 103 Credit: 14,948,929,771 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Hi, my 750Ti (ubuntu, driver 343.22) has been happily running GERARD long runs and NOELIA short runs WUs with decent crunching times. Now, when it comes to processing the new long runs NOELIAs, it does not go beyond 5-7% GPU utilisation (even with everything else stopped) and of course it would take forever to finalise the WU. My other system with 660Tis, ubuntu also and same driver is crunching them just right (88% GPU). Any idea? Thanks |
skgivenSend message Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Try suspending the task and resuming, restarting the client, restarting the system (cold start; shut down, power off, wait 1min). If it keeps doing this abort the task and try another. If need be, switch to long tasks. FAQ's HOW TO: - Opt out of Beta Tests - Ask for Help |
|
Send message Joined: 25 Mar 12 Posts: 103 Credit: 14,948,929,771 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Thanks skgiven, I've tried they all, also reseting and detaching but no luck. If I change to a short run WU or Einstein, the GPU utilization is OK but when it is a long run NOELIA it just stops using the GPU. So I will go back to short runs. |
|
Send message Joined: 28 Mar 09 Posts: 490 Credit: 11,739,145,728 RAC: 95,752 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
The NOELIA_ETQunboundx units do run a little slower than the GERARD_FXCXCL12_LIG_ units on my computers with GTX 690 cards, but nothing too dramatic. GPU usage is about 5% to 8% lower. Card temperature is about 5 degrees C lower. |
|
Send message Joined: 28 Jul 12 Posts: 819 Credit: 1,591,285,971 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Maybe something has changed on the Noelias? The NOELIA_ETQunboundx are running about 7 hours 25 minutes on my GTX 750 Tis, whereas the GERARD_FXCXCL12 take their leisurely 22 1/2 hours. https://www.gpugrid.net/results.php?hostid=223541&offset=0&show_names=1&state=0&appid= The Noelias do about 78% GPU load and 79% TDP, whereas the Gerards are about 84% GPU load and 89% TDP, as averaged by GPU-Z. (I am on Win7 64-bit, with the 352.86 drivers). |
|
Send message Joined: 6 Sep 10 Posts: 8 Credit: 3,494,747,495 RAC: 344,083 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Trotador I was having this problem also. Ubuntu 14.04 64 bit. I upgraded my Nvidia driver from 64-340.46 to 64-346.72 and now the long NOELIA_ETQunboundx are running normal. Jim |
|
Send message Joined: 25 Mar 12 Posts: 103 Credit: 14,948,929,771 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Trotador Thanks! I was figuring out that it would have to be the next thing to try but before I wanted to hear someone else in the same situation. Btw, you get good times with your 750Tis, overclocked? |
|
Send message Joined: 28 Jul 12 Posts: 819 Credit: 1,591,285,971 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
Btw, you get good times with your 750Tis, overclocked? I think you were referring to me. I am trying it at 1348 MHz GPU core clock, so that the GERARD_FXCXCL work comes in under 24 hours; they are usually about 22 1/2 hours now. I may not need any OC now, since it seems that they have gotten slightly shorter than they were originally, though whether that is by design or just how the work is going I don't know. |
|
Send message Joined: 6 Sep 10 Posts: 8 Credit: 3,494,747,495 RAC: 344,083 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
My 750Tis are stock Evga Geforce SC superclocked 02G-P4-3753-KR listed as core clock 1176Mhz, boost clock 1255Mhz boost clock. I have also found that the driver upgrade fixed problems I had on POEM@home, that I use as backup project. The 750Tis would not run there on the old drivers with out errors, now they run well. Jim |
|
Send message Joined: 25 Mar 12 Posts: 103 Credit: 14,948,929,771 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
My 750Ti is also an EVGA SC but I barely get half of the long run GERARD WUs under 24 hours even if the NVIDIA control panel shows 1320MHz. Both of you are achieving it routinely even in windows, mine is in Linux so I think there should be surely something I can improve. It is a 32 threads machine but I restrict CPU crunching to 30 of them. The CPU is running at 3.3 GHz which is lower than yours but not a lot. Then, what else? |
|
Send message Joined: 28 Jul 12 Posts: 819 Credit: 1,591,285,971 RAC: 0 Level ![]() Scientific publications ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]()
|
It is a 32 threads machine but I restrict CPU crunching to 30 of them. The CPU is running at 3.3 GHz which is lower than yours but not a lot. Linux should be a little faster than Windows, even at 3.3 GHz, which is not much different as you say from the 3.7 GHz on my i7-4770. But I don't know about 32 threads; that seems like a lot. Even if you leave two cores free, there may be other limitations, as on I/O depending on what other projects you are running. So I would try leaving one or two more cores free and see if that helps. There is also swan_sync as has been discussed, which ties a CPU core more closely to the GPU. But it is a bit different on Linux than on Windows so I am reluctant to offer specific advice; just search for it here on the forum. |
©2026 Universitat Pompeu Fabra