Message boards :
Number crunching :
Problem on GTX 560TI GPU
Message board moderation
Author | Message |
---|---|
Stegy Send message Joined: 14 May 99 Posts: 3 Credit: 1,336,203 RAC: 0 |
I have Gigabyte 560 Ti card. I did noticed, that several results are invalid (SETI@home Enhanced v6.10 (cuda_fermi)) http://setiathome.berkeley.edu/results.php?hostid=6238994&offset=0&show_names=0&state=4&appid= I did see, that several user have issues with Cuda. Is there any program, that can check if GPU (Cuda) calculations are OK? |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Hi Stegy, The current school of thought is that some of these factory overclocked 560ti's need some extra GPU core voltage for stability running Cuda, the theory being that the manufacturers may have spent the effort getting them 'game stable' but overlooked Cuda/PhysX stability. I would suggest to use something like nVidia Inspector or MSI Afterburner to put the core to 1.065Volts, and see if you get any further issues to look into after that. An alternative is to use the same program(s) to lower the clocks from the factory defaults. For Artefact Scanning & pushing the card I use OCCT on maximum complexity for 1 hour. zero artefacts under that amount of pressure for that duration is, IMO, truly stable enough. In the longer term I'm considering turning some of my Cuda test pieces that use Seti algorithms into some kind of stress test & bench, since some kernels have been shown to push fairly hard, though doing that right is a way off. In addition you may have better luck with newer application builds, plenty of help is available here if you consider it practical to keep an eye out for updates from time to time. [Edit:] beyond that, there is a whole evolving checklist of things to look it in the system. While I've had no problems running my own 560to out of the box at 900MHz, I do know of one with ongoing issues, and a few that have solved theirs with care. Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Stegy Send message Joined: 14 May 99 Posts: 3 Credit: 1,336,203 RAC: 0 |
I did run OCCT for GPU - shader complexity 8 for 2 hours. Test did not return any errors. Currently I have 10 valid and 71 invalid work units. So what is the problem (in the mean time I did also reinstall system). Is there bug in seti@home cuda fermi edition (6.10)? |
Slavac Send message Joined: 27 Apr 11 Posts: 1932 Credit: 17,952,639 RAC: 0 |
|
DMMD Send message Joined: 14 Feb 00 Posts: 118 Credit: 71,564,960 RAC: 0 |
Interestingly, the stock Vcore on the EVGA 560ti twin is 1.065 V. At that it is stable to the rigors of 41g. Very interestingly as well, is the decreased performance hit to using HT and NOT reserving logical or physical cores (resource contention) on this and the gtx 580 under the beta nvidia driver 290.36 and 41g. At least on the 560ti twin, the increased CPU load with the > 270 drivers is also not seen using Raistmer's AP masterpiece. I cannot speak to the thread safe issue under this config.....yet. The strategic use of HT et al, at task request time is NB.......... |
Tim Send message Joined: 19 May 99 Posts: 211 Credit: 278,575,259 RAC: 0 |
I have Gigabyte 560 Ti card. I use to have the same problem with one of my rigs. ID: 6022669 I have 2 gtx560tis. One with stock clock 823Mhz and one with manufactured over clocked to 900Mhz. The 900Mhz was sending invalid results. Not always but often. After many-many tests i downclock the 900Mhz to 823Mhz, and is running perfectly since then. Maybe you can do the same just to test the card. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
I have Gigabyte 560 Ti card. In the past months, I regurly encountered the GTX560Ti, unfortunatly by procusing errors. (Core-GPU-Frequency is quite high, 1.8GHz. compaired to other FERMI GPUs. The 470 runs at 1.201GHz, the 480 @ 1.400GHz. Ofcoarse you can OC them, but some app.s can/will crash, f.i. GPUgrid, already has a higher GPU load, but still needs a (fast)CPU. When i upped speed from 1.4G to 1.6GHz. the WU (Long runs-8 - 12 hours on fastest card), crashed. And on SETI, it was a (little) bit faster. Not noteworthy, though. (I'll add the AstroPulse NVIDIA app. rev521, now) |
Bree Send message Joined: 7 Feb 00 Posts: 9 Credit: 12,468,505 RAC: 0 |
Have a Gigabyte 950 it runs great as LONG as you don't let it down shift to a lower speed ie there set to down clock automatically. so when going from 950 to 600 then 300 then 50... the first time it shift it would throw a WU into error. if i didnt let it down shift then it was ok.. upshift seems to not bother it. just the down shift. had to make a rule as long as boinctasks is running and or boinc manager is running then DON'T down shift.. that worked pretty well . still get errors once and awhile but its from changing WU's during long no work periods. the card bothers me as it decided , running seti and its been on 5 mins. DOWNCLOCK to 50 mhz. also have talked to others that changed the parameters of the card to only allow the cards max speed at all times. only thing left on is shift in the power <autovolt and auto load sence> so you still get some savings vice a full on card all the time . that fixed it according to some others i have talked to . |
Slavac Send message Joined: 27 Apr 11 Posts: 1932 Credit: 17,952,639 RAC: 0 |
Have a Gigabyte 950 it runs great as LONG as you don't let it down shift to a lower speed ie there set to down clock automatically. so when going from 950 to 600 then 300 then 50... the first time it shift it would throw a WU into error. if i didnt let it down shift then it was ok.. upshift seems to not bother it. just the down shift. had to make a rule as long as boinctasks is running and or boinc manager is running then DON'T down shift.. that worked pretty well . still get errors once and awhile but its from changing WU's during long no work periods. Bump your voltage up slightly, that should solve the DC issues. Executive Director GPU Users Group Inc. - brad@gpuug.org |
Stegy Send message Joined: 14 May 99 Posts: 3 Credit: 1,336,203 RAC: 0 |
I did set voltage on my 560ti video card to 1.065V, but I still get invalid results in my results for CUDA. So what can I do, that I wont get any invalid entries any more. (OCCT also did not return any errors in stock voltage) |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
I suggest to downclock the core speed just to see if you get valid results. Thats what i did with a frineds card. With each crime and every kindness we birth our future. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
I did set voltage on my 560ti video card to 1.065V, but I still get invalid results in my results for CUDA. So what can I do, that I wont get any invalid entries any more. (OCCT also did not return any errors in stock voltage) Hi there, I've been working extensively with some members of the GPU Users group to nail these tricky cards down completely, and we believe we have them more or less mastered now (Thank's Slavac! :D). Mine's never missed a beat since installed, straight out of the box, so it's been a challenge ;) Our evolving checklist mostly involves the following for these cards, some of which will have already been covered. - Ensure your PSU has sufficient current on the +12 volt rail (At least 30 Amps per 560ti card, plus whatever the rest of you system needs peak) - If factory overclocked, raise the GPU Vcore (check) - physically reseat the card(s) & connectors. - Install later drivers, Use the 'clean install' advanced option (yours are old) - avoid applications not compatible with the newer drivers (i.e. stock cuda_fermi & beta OpenCL applications & other projects that have non threadsafe applications), your stock one predates 560ti by a year or more, try the Lunatics optimised one (x41g) which is significantly more robust & threadsafe for snoozing & exiting tasks without triggering weird issues. You'd want the 64 bit installer from Lunatics Windows Downloads - If errors still persist, there could be some driver caching old code still to go searching for & clean out, though the above approaches have solved some really weird issues already. Jason "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.