GTX 970 crashing

[grumpysculler]

Hi,

I recently upgraded my aging 5770 to an MSI GTX970 https://www.overclockers.co.uk/showproduct.php?prodid=GX-259-MS

Problem is, I get crashes with black screen and then the sound getting stuck on a loop. Unresponsive and it takes the power button to reset.

I've tried Inquisition and Titanfall, these crashes happen every single time between 10-20 minutes of playtime. I can run Furmark for at least 40 minutes without issue. I can run office apps all day without a problem.

I've cleaned out and reinstalled the drivers (using 344.75)
Dropped various quality settings in game
Dropped core and memory clock 100 MHz using MSI Afterburner

Temps are fine (max 70 on GPU, 50 on CPU), there's nothing obvious happening before the crash if I log sensor data. Nothing at all in event viewer.

Any suggestions, or is it RMA time?

Setup is:
i7-920 @2.67 GHz (not overclocked)
Gigabyte EX-UD5
6 gb RAM
650W Fractal Design PSU

Thanks in advance...

 

[grumpysculler]

Thought so, but always good to get a sanity check!

Thanks.

 

[grumpysculler]

Aaarghh...

RMA and then "no fault found".

A google on "gtx970 black screen crash" gives plenty of people with the same problem but no solutions.

My £300 paperweight is on its way back to me, I guess I'll have a muck about with settings and end up chucking it.

Any suggestions from the community?

 

[grumpysculler]

there was an issue with random x58 boards and 9 series , have you updated your bios ?

That's weird - I could swear I did a couple of years ago, but CPUZ says I'm running F7 (current is F13).

New BIOS incoming...

Try rolling back to an earlier driver?

Any suggestions as to which one might be stable?

I don't have the card back yet, but I'll do the BIOS now. Thanks for the suggestions.

 

[grumpysculler]

Hi grumpy,

I have the same CPU (OCed tho), ram and PSU as you, just different MB. I got a gtx970 a couple of weeks ago and I haven't had any errors; it looks like a MB/BIOS issue. I use the latest nVidia driver and it's completely fine.

Awesome - thanks.

I hope its the BIOS and not the MB itself...

 

[grumpysculler]

Bah blooming humbug.

So I have:
- card tested as OK by OCUK
- latest BIOS and drivers for mobo
- clean install of Win7 SP1 (fully updated)
- a couple of runs on memtest passed on the RAM
- used multiple versions of nvidia driver

And the only thing that has happened is that my reliable black screen crash has become a reliable BSOD. Code 116 which (quelle surprise) says there's an issue with the GPU. It happens much sooner, within 10 minutes of benching now.

I also had an odd issue where, running Heaven, one of the HDDs would constantly disconnect and reconnect. It's unplugged at the mo. Doesn't happen under HDD load, does happen with Heaven running. Could be a dodgy disk, could be dodgy MB, could be dodgy PSU?

The PSU should be OK, but I'm starting to wonder. Is there any way to check without just putting a meter on one of the ATX connectors while benchmarking?

If it isn't that, then I'm left with the MB. Either it's duff (seems unlikely) or incompatible and I really don't want to replace it. Aside from anything else, I went socket 1366 which probably means replacing more than just the board...

Help me OCUK, you're my only hope...

 

[grumpysculler]

Ensure there are no other things using the PCIe bus (e.g. sound card)

Ding ding ding I think we have a winner.

Up to an hour on heaven now and totally stable. HDD is still playing silly buggers from time to time, but I have a spare I'll swap in if I can't sort it some other way. 12V (single rail) looks stable with a meter on it but that won't catch any transients. Voltage doesn't move at all under load. It's 5 years old, but only working at half its rated power which should be plenty of margin.

The issue appears to be my USB3 card, which I completely forgot I had fitted. So small and tucked away at the top under the CPU cooler, never really think about it. Removed that just before the last heaven run and it's all tickety boo.

I'm going to leave it benching for another hour just to be sure.

All that bother over a £30 USB card. It will meet a suitably destructive (and theraputic) end.

Many festive thanks to all who chipped in. You've all been a great help.

 

[grumpysculler]

Well it BSOD'd (116 again) after 90 minutes of Heaven on extreme, but I can probably live with that. I'll see how long I get in DA:I or titanfall, but I rarely get over an hour at a time so might never see any problems.

4GHz (or near enough, I might keep it a bit cooler) has always been the plan. Just never got round to it. Kids happened instead...

edit: Better find myself a new USB 3 controller too. reading cards off my camcorder or DSLR takes an age on USB 2!

 

[grumpysculler]

Could be the PSU?

I'm still suspicious of it, but I don't really understand why it holds up to the load for so long before giving out (if that's the problem).

TDP isn't that much more than the 5770 which has been rock solid for years (145W vs 108W) so we're only talking 3 amps or so different. And a 650W PSU should really be capable of handling <350W, but then there is the funny business with the HDD.

The HDD could just be bad mojo - my regular work one started throwing SMART errors a couple of weeks ago and got replaced. And my other caddy just plain old died one night ("No disk found" in the morning). I could just be the harbinger of doom for hard disks...

I'm going to run memtest overnight just to rule out the RAM for sure and I'm getting some more tomorrow anyway, might try new sticks only first to see if it makes any difference.

If it isn't the RAM, then unless I've goosed the MB then it must be the PSU I suppose.

Are you able to try the card in another machine, or freinds pc?

Not really, although it's been back to overclockers on RMA and tested out fine on a variety of setups (they didn't say what).


scubes - BIOS update was trivially easy using Qflash. Put BIOS on memory stick, press key on boot, Hey Presto New BIOS.

 

[grumpysculler]

Fractal Design Newton R2 650W. 5 years old.

Superflower 650W is on this week only... Hmm...