Is my problem my GPU or PSU

[Magee_MC]

I have been having problems with running AOTS. Whether I'm playing the game or running the benchmark, less than a minute in my screen goes black and I have no way to recover the computer other than a hard power off/power on.

My system is:
I5 2500K at 4.3 on water
ASUS P8Z77 MB
32 GB G.Skil Sniper DDR3 1600
PowerColor R9 290 TurboDuo
Seasonic X650 Gold

What I have found is that if I turn the power limit to -25% in Crimson Global Overdrive, the problem goes away, so it's somehow a power issue. My temps never go above 75 on the GPU or 66 on the CPU, so that isn't the problem.

What I'm wondering is if this is a problem with my PSU under heavy load, or if it's a problem with my GPU. If it's the PSU, it's still under warranty, but how do I know that it's the problem. If it's the GPU, the same issue, how can I show that's the cause?

My system is rock solid stable except for AOTS and has passed every other test I've run on it.

Thoughts?

Thanks!

 

[xorbe]

I ran 2500K@4.5+Titan on a plain SS 550, that SS 650 Gold should be fine. BTW, my 2500K started getting flaky and I had to reduce clocks, before I replaced the whole setup. Other basic things to try,

1) Down-clock GPU / vram
2) Down-clock CPU
3) Remove 1/2 the ram (try 1 stick per channel)
4) Lower system ram speed / slower timings

 

[Magee_MC]

I ran 2500K@4.5+Titan on a plain SS 550, that SS 650 Gold should be fine. BTW, my 2500K started getting flaky and I had to reduce clocks, before I replaced the whole setup. Other basic things to try,

1) Down-clock GPU / vram
2) Down-clock CPU
3) Remove 1/2 the ram (try 1 stick per channel)
4) Lower system ram speed / slower timings

I tried a whole bunch of things trying to figure out the problem.

-GPU is running stock - if I have to underclock the GPU/Memory then there is a problem there. Should I test running it at the standard power level and down clocking?

-CPU - same problem with black screens whether I'm running stock at 3.3 or with my normal OC of 4.3.

-RAM - if the RAM is the problem, why would going to -25% on my GPU power resolve the problem?

-PSU - Is there any way for me to find out if it's unstable at a high power load?

 

[xorbe]

GPU is running stock - if I have to underclock the GPU/Memory then there is a problem there. Should I test running it at the standard power level and down clocking?

Bingo. It's the likely culprit.

PSUs are hard to test without swapping units ... it's how I wound up with an overkill SS-860W when my 2500K acted up.

 

[Magee_MC]

Bingo. It's the likely culprit.

PSUs are hard to test without swapping units ... it's how I wound up with an overkill SS-860W when my 2500K acted up.

Yeah, I completely understand. I've got an EVGA G2 850 watt gold PSU on order. I figured that worst case scenario, my SS 650 is good and I'll look at another build.

I'll give downclocking a try and see where that lands me.

Thanks!

 

[Magee_MC]

Ok, I just tried turning the clocks on my GPU to -25% and it went almost 2 minutes before crashing. I then tried -53% (the most I could turn it down) and it went through the benchmark fine.

It sounds like it's probably my GPU, but I wish that I could rule out it being my PSU under heavy load getting wonky. What bothers me is that it only happens under AOTS and my system is stable otherwise. I would have thought that if there was a problem with my GPU it would have shown itself before.

Is there anything else that I could try to isolate which one is the culprit other than pulling my PSU and trying the new one when it gets here?

 

[tweakboy]

Could be defective.. gl

 

[xorbe]

-25% and -50% are HUUUUGE drops. Did you try lowering vram instead? Maybe lower gpu clocks is just pushing on vram less. Throwing blind darts, haha.

One annoying thing is that the new generation of GPUs, when dropping clocks, the voltage drops too, so ... it's not as convenient when we could drop the clocks, but the voltage would stay the same, fixing minor stability issues.

 

[xthetenth]

Ok, I just tried turning the clocks on my GPU to -25% and it went almost 2 minutes before crashing. I then tried -53% (the most I could turn it down) and it went through the benchmark fine.

It sounds like it's probably my GPU, but I wish that I could rule out it being my PSU under heavy load getting wonky. What bothers me is that it only happens under AOTS and my system is stable otherwise. I would have thought that if there was a problem with my GPU it would have shown itself before.

Is there anything else that I could try to isolate which one is the culprit other than pulling my PSU and trying the new one when it gets here?

Do you have a way to see how much wattage you're pulling? You could take a look at your draw when running Ashes underclocked and when running something else known to be strenuous. If you're getting problems in Ashes when pulling less power than when running something else fine, I'd think it's the card. In that case I'd try running Ashes in DX11 and see if it's just the utilization somehow messing things up.

 

[KingFatty]

How is power getting to the card? Do you have a dedicated or separate rail providing the PCIe power plugs?

You could try using the standard molex-to-PCIe adapters, and connect them to a separate rail, to provide the direct power to the video card. That would at least isolate the specific rail of the PSU providing power to the PCIe plugs. But maybe your PSU has a single rail for everything? I'm assuming it doesn't and that you can try providing power using a different rail by using the power adapters.

 

[Magee_MC]

-25% and -50% are HUUUUGE drops. Did you try lowering vram instead? Maybe lower gpu clocks is just pushing on vram less. Throwing blind darts, haha.

One annoying thing is that the new generation of GPUs, when dropping clocks, the voltage drops too, so ... it's not as convenient when we could drop the clocks, but the voltage would stay the same, fixing minor stability issues.

You're right that the voltage drops when the clocks drop. Since it takes such a huge drop in clocks to get rid of the instability, my money is on it being the drop in voltage associated with the drop in clocks. I didn't try lowering the vram, but at this point I wouldn't think that's the problem. It definitely looks like a power issue since things that I do to drop the voltage removes the instability.

I'lll put the new PSU in tomorrow when it arrives and see what that gets me. If the problem is still there, then I'll start a RMA on the card.

 

[Magee_MC]

Do you have a way to see how much wattage you're pulling? You could take a look at your draw when running Ashes underclocked and when running something else known to be strenuous. If you're getting problems in Ashes when pulling less power than when running something else fine, I'd think it's the card. In that case I'd try running Ashes in DX11 and see if it's just the utilization somehow messing things up.

This happens in both DX11 and DX12. I ran a log on GPUz and when stable it's using for VDDC Power Out [W] <80 watts and up to 228 watts unstable at normal power levels.

 

[Magee_MC]

How is power getting to the card? Do you have a dedicated or separate rail providing the PCIe power plugs?

You could try using the standard molex-to-PCIe adapters, and connect them to a separate rail, to provide the direct power to the video card. That would at least isolate the specific rail of the PSU providing power to the PCIe plugs. But maybe your PSU has a single rail for everything? I'm assuming it doesn't and that you can try providing power using a different rail by using the power adapters.

It looks like the Seasonic X650 is single rail, so I don't think moving the plugs would work.

 

[Fizzy Coca-Cola]

I love your troubleshooting mind, magee_MC. I'm very happy after having read this. Are you having this problem in other demanding games? Dx12 affects CPU and GPU utilization, and I have been warning my friends that they need to get PSUs with much higher power ratings than their system currents needs. However, after a full thread read, these are my thoughts:

1) replace PSU and run test again. If you got the PSU on amazon, just return it if your system still fails with a higher watt, new psu.
2) Pull GPU and run the same test. Pull it out entirely. (I assume you did this already)
3) Put GPU in different slot (tests MOBO)
4) check cpu thermal paste, replace it
5) download newest DPU drivers or revert to previous GPU drivers
6) what GPU are you using?! I didnt see where you mentioned it. Some cards dont support DX12...

 

[Fizzy Coca-Cola]

Also, disconnect your CPU and motherboard and GPU connectors and reseat them, ensure good connections