Crash inside driver


We are in beta-testing with our upcoming game and experiencing crashes on nvidia cards, details are as follow:

Operating system: Windows7 x64 (Pro)
Drivers: 340.52 WHQL
Cards: GeForce GTX 780, GeForce GTX 760, GeForce GTX 880

(other test sets are unaffected GeForce 480, GeForce 660, GeForce 680 and GeForce 680M test sets works just fine)

The game engine uses DX10 as base rendering API.
The crash occures after 5 to 10 minutes of gameplay in random points, crash is always inside internal driver thread, the callstack is as follow:

5BBFF460 push ebp
5BBFF461 mov ebp,esp
5BBFF463 mov eax,dword ptr [ebp+10h]
5BBFF466 push ebx
5BBFF467 push esi
5BBFF468 push edi
5BBFF469 mov edi,dword ptr [eax+14h] → EAX == 0 here

Other threads do not process graphic related stuff at the moment of crash
Crash does not happend in debug builds (low fps)
Crash does not happend if the D3D10_CREATE_DEVICE_DEBUG flag is set
In about 30% situations ID3D10Texture2D->Map (STAGING, used to fetch GP-GPU FFT calculations) is near in timeline - but this is just a wild guess - commenting this out
does not make crashes gone, and freqency of them remains that same.
Validation and Debug of device do not indicate any criticals/erros
Game does not crash on ATI cards :] (or older nvidia models, only 7xx and 8xx series seems to be affected)

Any idea how to debug this kind of bug?

  • GeForce GTX 880 - was a typo, should be GeForce GTX 780M

BTW - i’v changed drivers all the way down to 320.18 (WHQL versions only)
the crass still occurs (inside nvwgf2um.dll - but the address is different, but the disassembly still looks that same (eax == NULL))