I never really noticed that I had stability issues with my GPU, because I've only used my PC for university related things and web browsing, until recently, when I decided to play some Diablo 3. I can't even finish an evening gaming session without crashing the PC even thogh Diablo 3 is not really the most GPU/CPU taxing game out there .
 
Here are my PC specifications:
 
Case: NZXT Phantom 820 with 7 fans
Mainboard: Gigabyte GA Z87X-UD7 TH @ Firmware Version 4
CPU: Intel i7 4790K Stock Settings (4.0 GHz - 4.4 GHz Turbo) @ 1.24 Vcore (I had 1.13 before, but I increased it in order to pinpoint this stability issue a bit easier)
CPU Cooler: Corsair H100i with 2 x Noctua NF-P12 fans
RAM: Corsair Dominator Platinum 4x8 GB 2133MHz @ 1.55 Volts
GPU: EVGA GTX 780 Ti Dual Classified Stock Settings with latest NVidida Drivers
PSU: Corsair AX1200i
 
I did some stress testing with the Intel's XTU and Aida64. I scored 1100 points in XTU and I had no problem passing the memory and cpu stress tests, and both tests lasted for at least 5 hours. Same things with Aida64: No problems at all with the stress test (it lasted 8 hours) if I only check the "Stress CPU + Stress FPU + Stress Cache + Stress System Memory" boxes. Once I add the "Stress GPU" option to it, the PC crashed after a couple of minutes. After the restart I only clicked the Stress GPU option while at the same time disengaging all the other options and the PC still crashed after a minute or so.
 
I also tried the Unigine Heaven Benchmark with extreme settings and I could only reach stage 22 at most, sometimes the PC crashed during stage 2 of 26. During the Benchmark, the FPS was constantly at 60, even right before the crashes occured. I checked the CPU & GPU temps with HWMonitor and the CPU temperature was 51°C and the GPU temperature 65°C. The GPU fan speed was at ca. 2100 rpm and the GPU clock speed was at 1150MHz @ 1162mV. I did no overclocking / fan curve modification whatsoever, left everything the way it was right out of the box, except for the bios switch, which I changed to the second position.
 
I also tried these same stress tests with the "load optimized defaults" bios option and it still produced the same results, with the exception of higher vcore settings and CPU temperatures.
 
Any advice on what I have to do to prevent these crashes? Should I provide more information on my PC system or maybe different stress test results?

Sajin
Sounds like the card isn't at fault. Sounds like the PSU has taken a dump. Does the crash act like someone just unplugged the power cable? If so, it's your PSU.

Well, yes, the crash occurs instantly, no Blue Screen of Death, just a black screen and a subsequent restart of the PC. Sometimes the PC takes a "break" and doesn't restart itself for a couple of minutes and sometimes it immediately restarts.
 
I'm not using the Corsair Link cable with the PSU, is that maybe the problem? I'm running the Corsair link software for the H100i Cooler, so maybe it expects the PSU to be connected, too? On the other hand, isn't that a good thing, that the PSU is not regulated by a volatile software program?
 
Is there any way to explicitly test the PSU performance? Should I disconnect and reconnect all PSU cables? Because I'm not sure what else there is to do, as I said, it's not controlled by the Corsair software...

Sajin
Best way to test the PSU would be with a multimeter, you could see if the +12v drops below the minimum ATX spec before the crash occurs. Yes, try reseating all the cables.

Fu**, I don't have a multimeter... I guess maybe connecting the PSU to the Corsair Link software might be another way to measure any voltage drops, no? Will do the reseating thing now...

Sajin

DarkDiver

Sajin
Best way to test the PSU would be with a multimeter, you could see if the +12v drops below the minimum ATX spec before the crash occurs. Yes, try reseating all the cables.

Fu**, I don't have a multimeter... I guess maybe connecting the PSU to the Corsair Link software might be another way to measure any voltage drops, no? Will do the reseating thing now...

Yes, you could use software to monitor the voltages. 

I'm back... I reseated all the cables and I've connected the PSU to the Corsair Link software... here are the results:
 

 

 

 
The first picture shows the normal PSU status with standard load. The next two pictures were taken during the Heaven Benchmark. The Benchmark went on for about 3-4 minutes, then the PC crashed... the AX1200i 24 A PIN ATX 12V box was jumping between 0 or 4.5/5/5.3. It was never constantly zero, but also never constantly 4 or 5. Is there any helpful info on these pictures that might help me to solve this problem?
 
Here is the Imgur album:
 
imgur dot com/a/RSXSr#0

Sajin
You don't have enough posts to be allowed to post links yet, 10 posts are needed, put spaces in the link to get them to show up.

can you see them under this link? imgur dot com/a/RSXSr#0

Sajin
It appears your +5v is out of spec, this could be due to monitoring with software though, that is why I recommend using a multimeter for checking the PSU, as software can be inaccurate. Is your PSU under warranty? If so, I would recommend RMA'ing it so you can get a newer one back to test with. 

You mean the AX1200i 5V V box? Top of the 4th row from the left? Isn't this one supposed to be just 5V? Or do you mean the box with the jumping values?

 
No, i'm talking about the hw monitor reading. +3.3 does not represent the PCI-E cable that connects to the GPU, +12v would represent that.

DarkDiver

Sajin
 
No, i'm talking about the hw monitor reading. +3.3 does not represent the PCI-E cable that connects to the GPU, +12v would represent that.

Is there any other way I can correct that +3.3 value? Maybe it's a faulty cable? Or is that something that I can't really change?

You can't change it. It's most likely incorrect anyways since you're using software to read it.

DarkDiver

Sajin

DarkDiver

Sajin
 
No, i'm talking about the hw monitor reading. +3.3 does not represent the PCI-E cable that connects to the GPU, +12v would represent that.

Is there any other way I can correct that +3.3 value? Maybe it's a faulty cable? Or is that something that I can't really change?

You can't change it. It's most likely incorrect anyways since you're using software to read it.

But let's assume that I buy a multimeter tomorrow and I get a correct +3.3 reading, what would be the problem then? Or would that correct +3.3 V reading imply that the PSU is not at fault here?

A correct +3.3v reading would mean the psu ins't having a problem maintaining that voltage on the +3.3v rail, so the +3.3v rail is within ATX spec.

DarkDiver
 
 
But are there more potential problems with my PSU readings besides these +3.3v? If there are no other problems with my PSU and the +3.3v rail functions as it should, then I won't need to RMA the PSU and I'd have to look for the problem elsewhere?

All three rails need to be checked +3.3v, +5v & +12v. If all rails pass with correct values then your PSU is most likely OK, but it's not a 100% guarantee that everything is OK, it just means your rails are maintaining their voltages correctly.

Sajin
If your PSU is under warranty I highly recommend you RMA it before buying a multimeter.

I rearragned all the cables on the PSU from the upper row to the lower row, and I immediately noticed that the red light on the GPU now turned green (the light is situated near the Motherboard IO port on the upper platine of the GPU). I guess the red light was an indicator to show that the PSU is not delivering enough power, not just to show that the GPU is working... Anyway, I did some quick Aida64 testing and it's going solidly now for about 20 minutes without any problems, but i'll leave it on for at least 12 hours to see if the system is really stable enough. The heaven benchmark was passed for the first time, also. This problem seems to be solved, however, some of the upper row PSU ports appear to be defective, so I guess it would still be a good idea to RMA it...

Sajin

DarkDiver

Sajin
If your PSU is under warranty I highly recommend you RMA it before buying a multimeter.

I rearragned all the cables on the PSU from the upper row to the lower row, and I immediately noticed that the red light on the GPU now turned green (the light is situated near the Motherboard IO port on the upper platine of the GPU). I guess the red light was an indicator to show that the PSU is not delivering enough power, not just to show that the GPU is working... Anyway, I did some quick Aida64 testing and it's going solidly now for about 20 minutes without any problems, but i'll leave it on for at least 12 hours to see if the system is really stable enough. The heaven benchmark was passed for the first time, also. This problem seems to be solved, however, some of the upper row PSU ports appear to be defective, so I guess it would still be a good idea to RMA it...

Good to hear. 

Thanks for your help, man! You saved me a lot of time. I was constantly thinking about CPU vcore and GPU temps, fan speeds etc., it never really dawned on me that it could actually the PSU's fault. So, thanks again! I will maybe pop in here again if Aida64 still encounters some problems, but i hope not

DarkDiver
I've been leaving the Aida64 stress test on for 10 hours without any problems or crashes. Here's a pic of HWMonitor and the Aida64 stress test after 10 hours of testing:
 

 
The only thing that's bothering me is the 1.788 vcore max setting. Is such a spike bound to occur during Aida64 or is it something that I still need to work on? I set the vcore to 1.15 manually and enabled the adaptive setting, because I've heard that the fixed setting might lower the lifespan of the CPU tremendously... but everything else, seems pretty normal, except maybe the +3.3v reading, right?

 
The vcore reading is most likely a false reading just like the +3.3v. I would use the fixed value so I wouldn't have to worry about spiking voltages on the CPU.

A high fixed value will deteriorate the CPU lifespan, 1.15v isn't very high, in fact 1.15v-1.18v is the stock value for your CPU. Temps look normal for your setup as the 4790k is a hot chip.