GPU tasks are causing Win10 machine to become unresponsive/restart (fixed by Disabling SLI)

[pandemonium]

Well, I've since had a few more instances of BAD_POOL_CALLER while running and the display driver attempting to reset after being woke and getting stuck. So, per recommendation of Retvari Zoltan on GPUGRID forums, I'm going to attempt to run these tasks with SLI disable, then removing the bridge if that doesn't work. If that's the solution, I may have to halt GPU tasks altogether, because I primarily game on this machine and doing that every time would be a pain.

 

[TennesseeTony]

Hopefully you won't need to remove the bridge, and the software switch will take care of the problem.

 

[Erich56]

on the same machine with two GTX980ti on which I have been crunching GPUGRID for 2 1/2 years on Windows XP, I recently installed Windows 10.

When crunching GPUGRID with these two GPUs, it happens every few hours that all of a sudden the monitor becomes black, because obviously the NVIDIA driver crashed.

All I can do is to push the off-button and make a reboot.

The Windows event log under "system" shows the warning "the graphic driver nvlddmkm does no longer react and was restored". This entry shows up exactly every 4 seconds from the time on the driver crashed. So, if between the time of the crash and the time I am back to the PC several hours later, this notice shows up a few thousand times.
Under "details" it shows an eventID 4101, and under event data "nvlddmkm"

What I did right away was a clean install of the latest driver 398.36, directly downloaded from NVIDIA, to replace version 388... which was installed along with Win10.
However, this did not help.

SLI is disabled.

Does anyone know about this problem, and could give me an advice how to solve it?

 

[lane42]

You could look here https://www.bing.com/search?q=what+is+nvlddmkm+error&FORM=R5FD

 

[StefanR5R]

What I did right away was a clean install of the latest driver 398.36, directly downloaded from NVIDIA, to replace version 388... which was installed along with Win10.
However, this did not help.

I don't have Windows 10 myself (only Win 10 and Linux), so I looked here instead what driver versions other people are using on Win 10:
http://www.gpugrid.net/performance.php
Users giller, jp de malo, and Woody have got driver 398.36 too, on Windows 10.00.17134.00 (but with Pascal GPUs, not Maxwell).

 

[Erich56]

honestly, I don't think it's the driver.
Because on my other two Windows 10 PCs I have 388... (the same that came with the new Windows installation here), and they work without any problem.

 

[StefanR5R]

Are there any remarkable differences between the crashing PC and the good ones?

Could it be that GPUGrid on Win 10 utilizes the two GTX 980Ti better than Win XP did, causing them to heat up themselves or the PC such that it becomes unstable, or drawing just so much more power that it becomes unstable? --- On the other hand, I recall GPU utilization by GPUGrid to be rather low, at least with 1080Ti and if only one task was run at a time, per GPU.

 

[Howdy]

When you say you did a "clean install" was through the nVidia installer or a program like this? I have used the program with some success, other times not so much.

 

[Erich56]

2

When you say you did a "clean install" was through the nVidia installer or a program like this? I have used the program with some success, other times not so much.

I removed the 388... driver with the DDU (which you linked), and then installed the 396.36 driver that I had downloaded directly from NVIDIA. So it was a "clean install".

 

[Erich56]

Are there any remarkable differences between the crashing PC and the good ones?

Could it be that GPUGrid on Win 10 utilizes the two GTX 980Ti better than Win XP did, causing them to heat up themselves or the PC such that it becomes unstable, or drawing just so much more power that it becomes unstable? --- On the other hand, I recall GPU utilization by GPUGrid to be rather low, at least with 1080Ti and if only one task was run at a time, per GPU.

In fact, Windows10 utilizes GPUGRID not as well as Windows XP does, because of the WDDM overhead which comes with all Windows OS from Vista on. This makes GPUGRID between 10% and 15% slower than in XP.

At any rate, I limit the GPU clock and thus also the temperature via the NVIDIA Inspector, having the GPUs run at 60/61°C.