PrimeGrid
Please visit donation page to help the project cover running costs for this month
1) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108910)
Posted 1378 days ago by Luke Formosa
I had changed the TDR delays under H_KEY_LOCAL_MACHINE \ System \ CurrentControlSet \ Control \ Graphics Drivers, it actually made the problem worse. When the graphics card crashes, the lag between mouse movements and pointer movement gets worse and worse, then at one point the mouse stops responding, the screen goes blank, and the graphics card fans loudly rev up to maximum speed. This lasts about 2 seconds, then screen turns back on, I get the "display driver has crashed" warning, and everything comes back to normal (except that all GPU tasks will have failed with a status of "computation error"). With a delay of 8 seconds it simply took 8 seconds of unresponsiveness for the computer to come back to life after the crash instead of the current ~2 sec.

Seems to be worse on my new (re)install, but I've actually done all the windows updates on this one versus not doing any updates on my last install.
2) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108428)
Posted 1412 days ago by Luke Formosa
It would be great if you added checkpointing/saving to sieve and AP27 apps like you have for Genefer.

Currently switched over to exclusively genefer. Have been running for a few days now with acceptable reliability. The way it recovers from driver errors is useful, and it also seems to throw up much less errors than the other projects.
3) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108120)
Posted 1426 days ago by Luke Formosa
There's also the TDP ("power limit") setting.


TDP stands for "Thermal Design Point". It's a measurement of the cooling ability of the device. People like to think of it as the max power limit, but it's not. It's the limitation of the cooling system.



I'm familiar with it - I used to be an R&D engineer at a semiconductor plant ;). In this context I meant the "power limit" slider on MSI Afterburner (and others) which presumably sets the max wattage (heat output) of the die as permitted by the boost clock. My 670 runs at 60-70% of full TDP when doing primegrid (as reported by GPU-z). Considering the GTX 670 has a TDP of 170W that means about 110W dissipated through the heatsink during crunching.

Are there any plans to update the apps? PPS Seive has been at v1.39 ever since I remember (almost 6 years and counting).
4) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108117)
Posted 1426 days ago by Luke Formosa
Update: Still no joy. Got a driver crash while running only Genefer at 900MHz core.

I'm at my wits' end. Looking for an alternative project to run. I think I will switch over to Collatz. Pity the 670s are 3x slower than 10-series cards on that project. Sigh.
5) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108116)
Posted 1426 days ago by Luke Formosa
Glad to be of assistance on this matter. Tbh for safety I do both core & memory, though you may want to test the different permutations. I literally calculate e.g 783MHz×0.9=705MHz.


Thanks :). In my case 980*0.9 = 882 so my 900MHz isn't far off. I left the memory at stock (6000MHz).

There's also the TDP ("power limit") setting. Doesn't seem to affect anything - with my 900MHz core clock, the actual GPU cores run at 1032MHz (GPU #0) and 1071MHz (GPU#1) while crunching Genefer.

I have to do more testing with PPS Seive as that seems to crash it most often.

What software do you use for the tweaking? At first I tried OC Guru as my cards are Gigabyte, but the UI was awful. Now I'm using MSI Afterburner and have it set to apply settings at boot though they seem to reset to stock a few hours in.
6) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108112)
Posted 1426 days ago by Luke Formosa
Hi Dave,

I've done some testing on your suggestion. Reducing the core clock from 980MHz down to 900MHz seems to have drastically improved the success rate.

Now, when you say a 10% underclock, what are you measuring exactly? Is that percentage of core clock, or %TDP? Do you adjust memory clock too?
7) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 108013)
Posted 1432 days ago by Luke Formosa
Yes, with SLI enabled.

However I used to get the same problems on this same PC before I added the second 670.
8) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 107955)
Posted 1434 days ago by Luke Formosa
I've found a 10% underclock helps 'solve' the issue, especially as hardware gets older


Thanks, sounds like a reasonable suggestion. Will implement and see how it goes, especially since my 670s are over 5 years old now :)

Tell BOINC not to use the GPU while the computer is in use.


I tried that previously, and although it works, I lose a lot of uptime because I frequently use the PC for a few seconds while boinc is running.
9) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 107913)
Posted 1436 days ago by Luke Formosa
That's true, Genefer didn't get a computation error after the driver crash. However I kept looking at the progress bar after the crash and it didn't increase any longer. On my PC the crash sometimes results in the GPUs not being seen any more (neither by BOINC nor by any other software, including the NVIDIA control panel) until I restart windows. So probably the reason for it to effectively stop running was because of the non-availability of the GPU.

The crashes occur much more frequently with Sieve. Actually coming to think of it, I may have been running Sieve and AP27 concurrently when the last few crashes happened since I have a dual GPU setup. So maybe it's always been PPS Sieve that's crashing it.

I suspect it's trying to put too much load on some component of the GPU (like maybe loading too much code in some register or some memory component), because if I leave the computer alone it runs mostly fine, but if I use the Chrome browser on this PC with PPS running I tend to get crashes every few minutes.

Any idea what I can do? With collatz conjecture, I could put a config file in the project folder that controls program settings (things like kernel sizes, lookup table sizes and so on). Certain extreme values of certain parameters resulted in immediate driver crashes. Perhaps Sieve is set too aggressively and results in crashes when other programs use certain GPU resources?
10) Message boards : Problems and Help : Frequent display driver crashes when running primegrid (Message 107908)
Posted 1436 days ago by Luke Formosa
Hardware: 2x Nvidia GTX670 in SLI
OS: Windows 7 Pro x64
Subproject: Mainly PPS Seive, but also AP26/27 and Genefer

When running primegrid, I get frequent display driver crashes. The screen goes blank for a few seconds, then comes back, accompanied by a notification that "the display driver has stopped responding and has recovered".

The status of any Seive tasks that were in progress when the crash happened becomes "computation error" so no points are granted.

The errors happen at random times, though it seems to happen more the harder I use the computer while Primegrid is running.

I have just finished completely reinstalling windows and the problem is still occuring. I can run GPUGrid, Collatz Conjecture and so on with no issues.

Any suggestions?


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 0.00, 0.00, 0.00
Generated 20 Apr 2021 | 11:42:47 UTC