Computer works fine with full-system stress, but crashes in less than 3 minutes ingame!

[GTMoraes]

Hey!

I just updated my GTX660 to a GTX970. Installation was a breeze, though Windows didn't detect the card at first (although I had a nVidia graphics card and drivers installed and running), but it wasn't much of a hassle.

I was wondering if my 3~4 yr old SevenTeam E-Force 550W would be able to power it, so I ran a system wide test before anything, so I've got the wPrime95 and EVGA OC Scanner and executed the GPU and GPU memory stress test and the wPrime95 CPU and RAM stress test.
My system ran them just fine with 99% GPU usage, 99% GPU mem usage, 95% CPU usage and 97% RAM usage for three hours straight, with the +12V link falling down to 11.8V (idles at 12.096V, pretty constant), which I guess it's acceptable, so I thought everything was fine!
Did a small overclock, just to get my OCD straight, of 121MHz+ and 110% power, so the GTX970 plays at 1500.8MHz, and 300MHz+ memory just.. because. Ran the tests for two hours and it was fine once again! +12V line didn't pass the 11.8V so the PSU is supporting it alright.

Fired up Far Cry 4, pumped up the graphics settings, started single player, went outside the safe house, and before I could finish my "Holy Sh..", the game crashed with a "device hung" message, and the little yellow exclamation icon near the system clock. Fired it up again and it couldn't get past the "press any key" part.
So I rebooted it. Fired FC4 up, went outside, ran for a while.. crash.
"Oh, it might be the game that's messed up. Been a long while since I don't play", and I went to play BF4.
Killed a guy, blew a tank, got killed... game crashed to a black screen and "Ding!" in the background. Alt-Tabbed it and it told me something about device removed, and the yellow exclamation icon was there.

I don't get it, what's so different in benches and games that makes games crash?? I didn't try to remove the overclock yet (yes that's stupid) but I'm guessing it will give me the same error!
GPU temperature doesn't get past 57ºC in any scenario. GPU temp doesn't go higher than 60

 

[ezskills]

Try running your game without the overclock and see what happens because it seems your overclock is not game stable only bench-marking stable.

 

[JackNaylorPE]

http://www.tomshardware.com/answers/id-2311121/power-supply-requirements-nvidia-gpus.html#14243229

11.8 volts is a 1.7% voltage drop.... well within the ATX spec of 5% but above the 1% I like to see for serious overclocking.

Which 970 ? MSI and Gigabyte shud have no problem w/ 1500... Asus Strix and EVGA will struggle.

P95 is useless for stability testing, especially in a post Haswell world. Nice for seeing how hot ya can get ya CPU, tho under any real world situation you will never be able to match it. You can be rock solid P95 stable and RoG Real Bench (which uses real life applications) will bring system to its knees.

As for GPU, you can be rock stable in every benchmrk out there but fail in game.... the benchmarks are very well coded and these poorly coded shove out the door ASAP games, much of which are console ports, will make things come tumbling down. For example, we have special Afterburner profiles that are used just for playing BF4

 

[GTMoraes]

http://www.tomshardware.com/answers/id-2311121/power-supply-requirements-nvidia-gpus.html#14243229

11.8 volts is a 1.7% voltage drop.... well within the ATX spec of 5% but above the 1% I like to see for serious overclocking.

Which 970 ? MSI and Gigabyte shud have no problem w/ 1500... Asus Strix and EVGA will struggle.

P95 is useless for stability testing, especially in a post Haswell world. Nice for seeing how hot ya can get ya CPU, tho under any real world situation you will never be able to match it. You can be rock solid P95 stable and RoG Real Bench (which uses real life applications) will bring system to its knees.

As for GPU, you can be rock stable in every benchmrk out there but fail in game.... the benchmarks are very well coded and these poorly coded shove out the door ASAP games, much of which are console ports, will make things come tumbling down. For example, we have special Afterburner profiles that are used just for playing BF4

Yeah, I didn't think 11.8V would be THAT bad, but who knows.. I was surprised, actually. That's an 3~4 years old PSU (I might say even 5 years old), which is running 24/7 since 2013, stopping only due to power faults. Does SevenTeam still makes PSU's?

It's an EVGA, GTX970 OC. I was about to get a 970 TwinFrozr, as it was cheaper, but thought EVGA was a better brand lol, bad deal.
Well, I've removed the overclock and managed to play half a BF4 match without issues. Re-applied the overclock partly (110% power limit + 121MHz, but not the memory overclock) and it finished the match successfully and was ready for another one (Ultra, no AA, 1080p, 150% image)
As for the Benchmark vs Games... I didn't know that. I thought processed code would be the same thing, be it a benchmark or a game. Weird

I'll have a go on FC4 to see if it crashes. If it does, I'll see what else can be done.
If it doesn't.. well, I'll be damned for being thread-happy and not trying my options before posting here =(

--------------

FC4 crashed, but that might be my fault. Game is modded since... well, weeks after release, and I didn't update the mods ever since, and the game received many updates since then, so I'll scratch FC4 for now.

You said something about RoG RealBench. I'm downloading it as I edit this message, and I'll have a go on it. Perhaps if everything goes okay there, it means it's fine? Or is it just "another benchmark"?

--------------
Weird. RealBench barely touched my GPU. The rest of the machine is alright, I know that.
Well, I'll see which demanding game I can fire up here

 

[GTMoraes]

bump

So I've reinstalled Windows so I could clear any driver issues off, updated it, installed the latest nVidia driver and went on playing.
Everything felt better and faster. That's fine. Managed to play BF4 hours straight, then proceeded to FarCry 4 (after removing the broken mod) and played for hours without issues.
I considered it as fine, until the driver died on me while I was working (just Google Docs basic stuff), what the hell?? It recovered fine, but it crashed nonetheless. Finished my work, went on to play FarCry 4 and it worked just fine. After it, I played L.A. Noire, and it crashed after 10 minutes...
??? The GPU wasn't even at 30% while I played!
Then I rebooted and played GTA IV... it crashed once again, after about 20 minutes... what's wrong this time around?! GPU clock is stock and the only thing custom is a more aggressive fan curve (to keep it cooler)

Is that a RMA case or a broken driver problem? I'm also worried that my PSU might not be supporting it anymore, but I've been logging and the lowest it hit was 11.568. It's within' ATX 5% as you said. Shouldn't it work?

---------------------

during a 3DMark bench loop, the driver crashed and 3DMark gave me this error log:

Unexpected error running tests.
Workload Single init returned error message: eva:: d3d11::rendering:: deferred_scene_renderer::render(): draw_medium_property_task for thread 0: File: device_context.cpp
Line: 818
Function: struct ID3D11CommandList *__cdecl eva:: d3d11:: deferred_device_context: do_finish_command_list(bool)

Expression: native()->FinishCommandList( restore_deferred_context_state, &result): DX11 call failed [-2005270522].

Device hung due to badly formed commands.
DXGI_ERROR_DEVICE_HUNG: ID3D11DeviceContext::FinishCommandList: