GTX 650 issues

[Thunderclap]

So this wasn't as often before, but now it's every 5 god damn minutes. My kernel drivers keep crashing, and I've got literally 0 clues why.
On GPUz it says "util" under perfcap reason, so I looked it up and people said it's the PSU - as in the GPU is too power hungry for my PSU, but that is definitely not the case.
A 450w Antec PSU is perfectly fine for a GTX 650 that doesn't even use a 6 pin connector, and a quad core 45W CPU.
And it never bloody stops, I mean LITERALLY, every 5 minutes I get a blackscreen for a few seconds and then it recovers.
It's not the CPU OC I made, that works perfectly fine - even returning to stock doesn't help.
I've tried fiddling with the voltages and clock speeds, but nothing there either.

Mind it, this happened with multiple drivers.

 

[Frick]

What are your temps? Depending on the answer I'd say the GPU is on the fritz.

Util I think means it just isn't used more.

BTW, does is happen when it's idling as well? Also, ehat OS are you on?

 

[Thunderclap]

So GPU *never* passes 55°C.
PC is rarely idle, but it does happen extremely rarely during lighter loads.
Wondows 7.

 

[Toothless]

Your issue might be Wondows.

 

[SomeOne99h]

Tried reseating the GPU?
Tried another PCI-Express slot?
Tried to use another PCI-Express cable?
Can you try to use your video card in another system? like your friends\family member?

If you tried in another system and it works fine on that other system, then PSU is not okay.
The PCI-E slot might be not okay or needs cleaning from dust. If everything checked, maybe the motherboard is acting funny.

 

[newtekie1]

Util under perf cap reason means that the GPU is limited by utilization. In other words, there isn't enough work being sent to the GPU for it t be fully used. This is normal when you aren't running something using the GPU.

My guess is the GPU is failing, and is unstable, which is causing the driver to crash.

 

[Thunderclap]

Reinstalled windows multiple times, changed bugger all.

Tried re-seating it, nothing.
My mobo is ancient, do you really think it has another PCIe slot?
And as I said, it doesn't even use a 6 pin connector.
And with my other GPU, it works fine. PSU is fine, it's a brand new PSU.

Yeah, I see that during games.

 

[Toothless]

You try upping the voltage a little / downclocking the card?

 

[Thunderclap]

"I've tried fiddling with the voltages and clock speeds, but nothing there either."

dude. I did it already.

EDIT: Ok so, I managed to fix it yesterday, I didn't post it here because I didn't want to jump to conclusions, but now again, it doesn't work. Now I got no fucking idea why it won't work, and it's pissing me off immensely.

 

[FR@NK]

Might be time to replace it.

 

[Frick]

"I've tried fiddling with the voltages and clock speeds, but nothing there either."

dude. I did it already.

EDIT: Ok so, I managed to fix it yesterday, I didn't post it here because I didn't want to jump to conclusions, but now again, it doesn't work. Now I got no fucking idea why it won't work, and it's pissing me off immensely.

What did you do to seemingly fix it?

Last effort is to try it in a different machine.

 

[Thunderclap]

Don't have the $ to replace it. If I did, I wouldn't bother with this half broken piece of sh*t.

What fixed it is deleting any trace of nVidia off my PC, installing oldest drivers I could find and updating to current ones from there. For some damned reason it worked.

 

[SnakeDoctor]

=video memory stress test
http://www.majorgeeks.com/files/details/video_memory_stress_test.html

 

[Thunderclap]

"[17.10.2016. 14:48:15] Test started for "Primary Display Driver (NVIDIA GeForce GTX 650)"...
Trying 16bpp RGB:565 mode...OK
Trying 16bpp RGB:555 mode...OK
Trying 16bpp BGR:565 mode...NOT SUPPORTED (Code: 80004001)
Trying 32bpp RGB:888 mode...OK
Trying 32bpp BGR:888 mode...NOT SUPPORTED (Code: 80004001)
[17.10.2016. 15:17:48] Pass completed (0 errors found)."

so yeah.

EDIT: Lol, now I ran it with "use onscreen buffer", and tell me, is having 5.9k errors good? Keep in mind, the number's climbing.

Error looks like this:
"Error at [3C7E9240]: must be FFFFFFFF, but found 00F0F0F0 (bits: 11111111000011110000111100001111)"

EDIT#2: Did an OCCT test, crashed twice during 1 min test.
Tested another few times. Only apparently happens with DX11, but not 100% sure. Found 9 errors, doesn't tell me anythin' about them.

 

[xorbe]

Card is failing, you're getting a TDR every 5 minutes. Last ditch, try a fresh Windows install to rule out bad driver or malware (yeah they cause TDR too.)

 

[Thunderclap]

Reinstalled windows a few times already.

 

[Yukikaze]

Looks like your VRAM (or the card's memory controller?) is going bad. The card is dying.

 

[eidairaman1]

dust everything out, reseat the card. Last but not least replace the card

 

[Thunderclap]

dust everything out, reseat the card. Last but not least replace the card

Did it already, doesn't hurt to do it again I guess. Last part tho, as I said, an average student like me can't afford a new GPU. When smth like R7 250 costs 120€+, and I pay my apartment like 150€ a month, I wouldn't like living on the street with a brand new GPU to get stolen or smth

 

[Evo85]

Check voltages going in to the card at the connector. Watch for volt spikes when the card crashes. If the voltages are solid, you either have a bad card, or a bad mobo.

I had a similar problem a few years back. It turned out to be the PCI-E socket was bad. That too was a single socket board so I had to purchase a new one. But I had zero problems after that.

 

[Frick]

Time for some cooking. If you dare. I've done it twice, one on a laptop motherboard and once on a GPU, both times saved the day. Do it wrong and it'll be completely ruined. But it's not difficult, just be careful.

BTW, can you try it in a different system?

Also you can keep an eye out on the used market, you get a lot of card for little money these days, depending on where you live. What games are you playing?

 

[Thunderclap]

I'm not playing anything too demanding - Neverwinter, Trackmania Nations Forever and DotA2 every now and again when my buddies invite me.
Here, an HD 4870 costs like 50-70€ so yeah. Expensive poor country named Croatia. I'd put it in the oven, but I don't wanna risk it. I've done some of my own fixes now, because the GPU was pretty much dead a few minutes ago, we'll see if they work, and I'll tell ye when it's ready.

Don't have a different system to try it on :/

 

[SomeOne99h]

Check voltages going in to the card at the connector. Watch for volt spikes when the card crashes. If the voltages are solid, you either have a bad card, or a bad mobo.

I had a similar problem a few years back. It turned out to be the PCI-E socket was bad. That too was a single socket board so I had to purchase a new one. But I had zero problems after that.

Evo85, I tracked your old thread https://www.techpowerup.com/forums/threads/games-crashing-not-sure-what-to-do-next.66757/
He has kernel drivers and bluescreens whereas you had read errors during gaming and without bluescreens.
Which means his (the OP's) PCI-E slot is fine? I guess his video card is bad.

 

[Thunderclap]

Yep. It was the GPU. Did smth similar to baking it, and now its kinda working, textures look a bit more janky in trackmania for instance, but it works. Question is for how long, but by that time I hopefully will get some money or at least sell a kidney kappa

 

[eidairaman1]

No telling. Hopefully you dont get a short in it that takes the rest of the rig with it.