NVIDIA driver crashes for a specific game

[yoadknux]

Hello,

I've been playing Overwatch on a setup of 8GB 1866MHz DDR3 (Corsair Vengeance), i7 2600k @ 4.5GHz, and a very heavily overclocked MSI 970 GTX. The PSU is a Seasonic X-Series 650W, and the OS is Windows 10. The game has been working perfectly fine for a few months, and never crashed after many many hours put into the game.

Last week I upgraded my 970 to a (Gigabyte Windforce) 1080. Ever since then, I cannot play Overwatch for more than 1-6 minutes before the game crashes. I'll be more specific - The OS doesn't crash, it's Overwatch that crashes and leaves behind a message "Your rendering device has been lost", which after talking with Blizzard's support is related to the driver. I've only noticed this crash on Overwatch and not on other games.

Here's what I've tried thus far in order to isolate the problem:
1. Clean windows 10 install - still crashes.
2. Remove the CPU OC - crashes.
3. Run Valley benchmark for a few hours, FurMark, OCCT error detection - Nothing, no crashes.
4. Run Prime95 Blend - CPU is hot, within the 70-78 range, but no crashes.
5. Run different games - The Witcher 3, Heroes of the storm, etc - No crashes, no black screens, games run perfect. I could get momentary FPS drops, though it's probably normal.

Here's what I want to do:
1. Replicate the crash on some other application. I'm looking for the most stressing GPU/CPU/Ram tests. Is there anything I can run to stress all three at once?
2. Maybe the driver actually crashes on other applications, but just manages to recover through it instead of crashing the application? Is there some way to view a log of the NVIDIA driver crashes?
3. I want to understand whether the GPU is faulty, or some other hardware component, or maybe the driver has some conflict with other drivers or services. If the GPU is faulty, I'll use my warranty. If some other hardware component, I've been wanting to upgrade the socket anyway. But I want to be sure before I do anything.

Any tips/insight?

Update: just to clarify the issues I'm having - On my new 1080 GTX, Overwatch crashes, and it's specific to this GPU. My 970 never crashed OW, same OS same build same drivers. I tried taking my 1080 to a different PC, fresh install - exact same crash. Additionally, I experience artifacts on Crysis 3, as shown in a video in page 1. However, the GPU passes all stress tests out there (listed them on page 2) - for hours. It runs heavy, demanding games such as Witcher 3 and Deus flawlessly. I'm trying to understand if this is a GPU issue or not. If it's an hardware issue, I'd be very interested in knowing what exactly is going on - for science!


Thanks!

Edit: Problem solved by replacing the Gigabyte 1080 with MSI 1080. The Gigabyte GPU is not faulty, I've checked myself with every single stress test I could get my hand on, also sent it to a lab which made sure it's fine. Crysis has built in artifacts, and Overwatch is very picky for some reason...

 

[SPL Tech]

If the driver crashes and recovers, Windows will let you know. Further, if it happens you will notice. The screen will go blank for a second or something will happen, it's not a seamless process. You can try Realbench. That will max out both the GPU and CPU, as well as much RAM as you want (you select how much RAM you want it to use). You could also just run two apps at once, like Heaven Benchmark with Prime95 running at the same time.

 

[yoadknux]

If the driver crashes and recovers, Windows will let you know. Further, if it happens you will notice. The screen will go blank for a second or something will happen, it's not a seamless process. You can try Realbench. That will max out both the GPU and CPU, as well as much RAM as you want (you select how much RAM you want it to use). You could also just run two apps at once, like Heaven Benchmark with Prime95 running at the same time.

Hi, thanks for the reply!

I went on ahead and downloaded RealBench. I've performed 3 tests. For the first one LuxMark crashed within 10 minutes or so. For the second test I reduced the CPU to 4.2GHz and RAM to 1600MHz, crashed within 10-20 minutes again. For the third one I disabled MSI afterburner and the windows screensaver (people claim they interfere) and the stress test passed with no errors for 8 hours. There was no OC whatsoever on the MSI afterburner, I only disabled the interface.

I think I shall repeat the test again to make sure I didn't fluke it, but I think if I pass it for 8 hours my hardware components should be pretty much OK. What do you think?

Edit 1: RAM is back at 1866, passed 30 minutes. Will perform a 2 hour test soon.

Edit 2: With RAM still at 1866, passed 2 hours.

So in total, I've passed an 8 hour test, 30 minute test, and 2 hours test.

 

[SPL Tech]

Sounds like it's stable. I only run benchmarks for about 30 minutes. That's stable enough for me and then I just start gaming. If the game crashes, I drop the overclock a bit more.
Does the display driver still crash in the game?

 

[yoadknux]

Sounds like it's stable. I only run benchmarks for about 30 minutes. That's stable enough for me and then I just start gaming. If the game crashes, I drop the overclock a bit more.
Does the display driver still crash in the game?

Yes, it crashed just now within a minute of gameplay. Weird thing is that in all those crashes, not once I've seen the driver crash message from windows. I get "Your rendering device has been lost" by Overwatch and the game crashes, but I don't get any notification about the driver crashing.

I just don't understand this behavior. I've used this exact gaming system (with 970 instead of 1080) and it worked perfectly. Now I have a 1080 and it crashes Overwatch, but only Overwatch crashes. No matter what other game I try to play, or which benchmark or stress test I try to run, it doesn't crash. The Valley benchmark result of this GPU looks normal.

I just don't really understand if this is an hardware issue or not. I'd think the GPU is maybe bad because the 970 worked, but this is the only application that crashes. I don't really know what to think right now.

 

[EarthDog]

Ehhh, swap drivers.

 

[Bishoff]

Yes, it crashed just now within a minute of gameplay. Weird thing is that in all those crashes, not once I've seen the driver crash message from windows. I get "Your rendering device has been lost" by Overwatch and the game crashes, but I don't get any notification about the driver crashing.

I just don't understand this behavior. I've used this exact gaming system (with 970 instead of 1080) and it worked perfectly. Now I have a 1080 and it crashes Overwatch, but only Overwatch crashes. No matter what other game I try to play, or which benchmark or stress test I try to run, it doesn't crash. The Valley benchmark result of this GPU looks normal.

I just don't really understand if this is an hardware issue or not. I'd think the GPU is maybe bad because the 970 worked, but this is the only application that crashes. I don't really know what to think right now.

As ED said, try another driver. Current NV drivers in the 380's have issues. Try the 378 version as most people have been rolling back to that one due to Windows 10 CU. You may need to go even further back, but definitely try another driver before blaming hardware.

 

[yoadknux]

I tried rolling back to 362 or so... Nothing. Same error, and again, only on ow. It's crazy, I've even tried overclocking it to get it to crash on Valley and it didn't crash. Maybe the card isn't seated properly or something? Should I try to plug it into a different PCI-E slot?

 

[SPL Tech]

You could try underclocking the card. Dial back the core and RAM as low as your tuning program will allow, set the fans to manual 100% speed and then try the game. You can try a different PCi-e slot.

 

[yoadknux]

You could try underclocking the card. Dial back the core and RAM as low as your tuning program will allow, set the fans to manual 100% speed and then try the game. You can try a different PCi-e slot.

I'll try underclocking and playing the game. I guess the fan is high to prevent possible overheating?

To be honest I'm more interested in understanding the situation than playing that specific game. I really am trying to understand if I just got a bad GPU. I tried to make it crash on anything else but couldn't. I even overclocked it today and ran Valley for 4 hours and the bench came out just fine. Maybe Overwatch detects faulty hardware that neither FurMark, Valley, OCCT, Realbench and Witcher 3 can detect... Or maybe OW is badly written. Don't know. That's what I'd like to find out.

Any other suggestions for GPU stress tests?

 

[yoadknux]

Sorry for the double post. Wanted to update something that may have worked, but still leaves me puzzled. I've done 2 things: 1. Play around with the Windows TDR settings, and 2. underclock my card as much as MSI afterburner allowed. The game hasn't crashed... Yet. So heavy underclock seems to work.

BUT...

I just don't understand what's going on.

Let me sum up everything I tried, and maybe someone will know to identify the problem.

The problem: Overwatch crashes GPU drivers quickly. I get "Your rendering device has been lost" from OW and "Overwatch has been blocked from accessing graphics hardware" from Windows. Here's everything I tried:

Evidence for GPU problem:
- Go back to my 970 - Works.
- Underclock the 1080 and modifying TDR - (Probably) works.
- Clean Windows 10 install - failed.
- Placing my 1080 at a totally different PC, fresh install - failed.
- Other drivers - failed.
- Other PCI-e slot - failed.

Everything points out to the fact I've received a faulty GPU... or is it? I was unable to crash any other application out there.
- Valley 4 hours - Stable
- Furmark 1 hour - Stable
- RealBench 1 hour - Stable
- OCCT 1 hour - Stable
- Plenty of hours on Witcher 3 and Heroes of the storm - Stable
- Artifacts/stuttering - None
- OW forum flooded with this error - but no particular cause or solution.
- Overclocked stability - Surprisingly high. I was able to hit over 2k core with no issues... left valley overnight, no crash, even though we're talking 200+ core.

I could use the warranty on my 1080 GTX, but I want to be sure that this is an hardware issue, specifically faulty GPU. However, I am not convinced that I received a bad GPU. Any tips guys? I've pretty much hit a dead end. It's not that I care that much about OW, I care about understanding this behavior.

 

[SPL Tech]

Well if you have the extra cash you could order an identical 1080 and when it comes return the defective one that way you can try replacing it without having to be without a card forever.

 

[yoadknux]

Well if you have the extra cash you could order an identical 1080 and when it comes return the defective one that way you can try replacing it without having to be without a card forever.

Sadly don't have that extra cash, haha

I tried playing on some other games now. I went with Battlefield 1, Deus Ex Mandkind Divided, and Crysis 3. The first two ran perfect, just like intended. The last one... ran ****. At first I saw little GPU usage and massive FPS drops (down to ~20-30). I then started seeing artifacts, as shown in this video:

Artifacts are clearly visible. On the top left line you can see the hardware specifications. The top line is the GPU temperature, core & memory speed, and usage. Note the GPU is not overclocked. The line below is for the CPU: usage and temperatures.

I tried getting artifacts on other stress tests, still nothing. Valley, FurMark, Kombustor, RealBench, OCCT, GpuTest... hours on hours and nothing.

Makes me wonder if Overwatch and Crysis can detect something that other games and applications cannot. Someone once suggested that it may be a RAM issue. Can RAM cause artifacts, or are those artifacts GPU related?

 

[SPL Tech]

hmm that's starting to look like a hardware issue. I am not aware of any software or driver issues that can cause artifacts--I think that's exclusively hardware problems and usually from overclocked memory. I think the next step would be to try the GPU in a different computer. Maybe ask a friend to swap cards for a bit? It's looking to be either a GPU problem or a motherboard issue, and I would lean more toward GPU. Do you have another PCI-E slot you can try it in? If you do and the card has the same problems, I'd RMA it.

Try running that same scene in Crysis but with the card's core and memory underclocked as far as possible. I think it should go to -500 on the RAM. If the issue goes away, you have your answer as to what's the problem.

 

[EarthDog]

One game showing artifacts does not make a faulty card..

 

[SPL Tech]

One game showing artifacts does not make a faulty card..

I am not aware of any software issue that can cause artifacting. It's pretty much always a hardware related issue, or more commonly a clock speed issue. usually something with the GPU memory.

 

[EarthDog]

One game showing artifacts does not make a faulty card..

 

[Kenrou]

One game showing artifacts does not make a faulty card.

This. If a card is faulty it will show in all games or at least in the "heavier" games, the ones that tax it the most, never just one. If it passes Heaven/Furmark that will drive any card to it's proverbial knees the hardware in theory is sound and working as intended. Crappy drivers or not installed/uninstalled properly will cause artefacts, i have had that happen to me a couple times in demanding games like CoD Black Ops 3/Advanced Warfare that use an insane amount of VRam.

While i wouldn't know if it's proof of concept or not but while OP does have artefacts in Crysis 3 the game is running smooth/fast as it should be and the recording program doesn't seem to have had any problems.

 

[SPL Tech]

It's not one program, it's two and two makes a pattern. He said he's having driver crashes. Hardware issues can cause different problems in different games. Yes, while in theory a hardware issue should translate to issues in all games, but the problem is computers are impossibly complicated and theory never really translates to reality. Things that theoretically cant happen do happen all the time. If it's not hardware then what is it? He already tried rolling the drivers and reinstalling the OS. That would solve any corrupt software issues or corrupt .dll issues.

 

[EarthDog]

I thought it was overwatch only. Realbench is for the cpu and is another issue.
Edit: i see.. crysis..

Did you try reinstalling overwatch/crysis? I see windows, but was the game(s) on a different drive??