My GPU wasn't dying, my PSU cable was

Posted by Remarkable_Mall_793@reddit | buildapc | View on Reddit | 30 comments

For like 3 months I was convinced my RX 7900 XT was cooked. Random frame drops, stutters, occasional black screen in heavy games, the whole thing. Ran benchmarks, temps were fine, drivers reinstalled like 4 times, even reseated the card twice.

I had some money saved on the side and was about to pull the trigger on a replacement, maybe step down to a 4070 Super or something.

Turns out it was the PCIe power cable. I was using a single 16 pin adapter that came with my Corsair RM1000x and at some point the connection on the GPU end got slightly loose. Not fully loose, like it looked fine, but apparently just enough to cause power delivery issues under load.

Swapped it out for the other cable that came in the box and its been 3 weeks with zero issues.

What kills me is how long I spent looking at GPU temps, VRAM usage, driver logs and never once thought to check something that basic. The cable looked seated fine visually, thats what got me.

If you have a high end GPU and youre getting weird load-only instability, check your power connectors before anything else. Would have saved me months of stress lol

[-]

asianfatboy@reddit

I should check mine because I still keep getting random reboots even after changing graphics card. It's hard to tell if it happens during massive GPU load. I'm not playing very demanding games at this point. Snowrunner, Insurgency Sandstorm, RDR2. From 5700XT to 9060XT. That or maybe I need a fresh new install of Windows 10 at this point.

I use 2 separate pins from the PSU to power my card though. Maybe the PSU itself is bad at this point. It's a Seasonic Focus Gold 650W that's maybe more than 4 yrs old.

[-]

CheesyItalian@reddit

obligatory maybe you should re-paste the GPU?

[-]

asianfatboy@reddit

Still a fairly new 9060xt. Unless asrock does a shit job of making their cards, doubt that's the issue. Temps are even better than on my 5700xt.

[-]

blind616@reddit

This is similar to my scenario. My system crashing isn't so random but more like on GPU power spikes. Any game I load that moderately uses the GPU crashes the whole system. Temps are fine etc (60~70ºC).

The weird part is that furmark can stress the rig easily for 10 minutes without any crash, running the GPU at 350W, maybe because the load is constant and progressive, but it's weird how stable it is on Furmark.

[-]

asianfatboy@reddit

Have you done any troubleshooting/testing to narrow it down as best you can?

Weirdest thing is Snowrunner is the worst offender as it reboots my PC very frequently after 2-3hrs. Sometimes I get a BSOD. Insurgency isn't as frequent and sometimes only the game crashes. RDR2 does not force a reboot but only CTDs. I frequently play games while watching youtube livestreams too on my 2nd monitor and there's this weird glitch where the video pane has sections flicker all over the 2nd monitor, even over the areas where the video isn't position, like over the chatbox. All of these also happened when I still had my 5700XT. I sold that card and guy never called me about any random reboots so it's something with my system.

I'll try to do a clean install of Windows once I move files from my primary SSD to another. See if some weird apps/programs left behind is causing something.

Shot in the dark, it might also be RAM related. I have a 4x8GB DDR4 3600mhz Corsair LPX setup. But I'd realistically can afford a PSU replacement vs. getting a 2x16GB DDR4 3600mhz kit at the same price of a 650-750w PSU lol.

[-]

zeddsworth@reddit

It could be your motherboard. Google your mobo model and see if other people have issues with it.

[-]

blind616@reddit

Yes, I did some tests yesterday. here's the information I gathered (some of it I've mentioned before):

Furmark can run for over 10 minutes, GPU (rtx 3080 10gb) would sustain 350W easily. The CPU (ryzen 5600x) never goes above 76W.
Two games I tried would crash almost immediately, on the main menu most times.
Due to the symptoms looking like hardware-related and nothing relevant on the event logs other than a "unexpected shutdown" event, I did not try changing the OS yet (resetting windows/using Linux)
Both games were ran through Steam.
When the system turns off, the temperatures are fine (60~70ºC on most components, none above 90ºC). Furmark would run the components to nearly 100ºC and was stable.
Voltages are fine, they're at recommended values and have been the same configuration for nearly 4 years.
Undervolting didn't bring much luck. I tried setting the power to 50% and flattening the curve starting at 0.9V, same issue.
As for RAM, we have similar setups. I can rule that one out easily, since I upgraded yesterday from 2x8gb to 2x16 gb. Both Corsair DDR4 3600mhz, specifically (CMK16GX4M2D3600C18 -> CMK32GX4M2D3600C18)

[-]

asianfatboy@reddit

gonna try your suggestions with the RAM, thanks!

TBNRhash@reddit

Just clarifying, you didn't swap out the actual cable that connected to the PSU with another one, or initially connected a cable to the PSU that didn't come with it, right? I didn't understand your post properly, sorry!

TetePepeF@reddit

he has a fully modular psu i’m assuming, so he had a backup pcie

AYasin@reddit

my Corsair RM1000x

Said PSU is modular.

Swapped it out for the other cable that came in the box

Adding these together answers your question.

Only addresses half of it. I wasn't sure if the cable they originally used came in the power supply box, the second quote is a bit ambiguous in that sense, even with the use of "other".

scientifichooligans@reddit

It sounds like he used another pcie cable included with the power supply. OP mentions he grabbed it from the psu box

WorseElk@reddit

Should have swapped the cable before dropping cash on a new GPU, that's always step one when weird artifacts start happening under load.

Fredasa@reddit

One wonders whether there would be any way of diagnosing this kind of issue, so that the best path forward wouldn't default to: arbitrarily replacing parts until thing works again.

Smagjus@reddit

Some issues are a nightmare to debug. I had infrequent GPU driver crashes for a while. In the end it was a slightly unstable CPU undervolt that was the culprit. The 9800X3D did not undervolt as well as I thought it would.

The hardest thing I ever successfully debugged was a massive stutter I would get in two of the Yakuza games. The only way I pinned down what was going on was by systematically disabling every single feature of Windows and the BIOS.

The culprit? Ethernet. The disable Ethernet and the stutter vanishes.

🤪

Having similar issues on my PC probably with a similar cause, but the symptoms are different. In my case the whole thing shuts off when i run a moderately intensive game.

What annoys me the most is that running furmark for 10 minutes works fine, but other games crash at the main menu screen or as soon as i load a save.

In my case the whole thing shuts off when i run a moderately intensive game.

That sounds way familiar, and unfortunately I feel like there's never been a catch-all answer to it. Best I can do:

Furmark may not be testing the entirety of VRAM the way a game potentially can. Could be worth looking into.

Possible factory OC that trips up the GPU during transient spikes while gaming but not the stress test? Mentioning this because I've read some folks having luck under-volting.

Actually didn't check how much vram was being used in furmark, but other vram intensive applications would run fine.

I've also tried several settings in undervolting without much luck.

prismstein@reddit

Got the reverse for my 4080S: Random display problems and hangs, cable plugged in snugly, temps and stuff all normal, reinstalled windows etc, turns out the 16pin cable that came with my CM V1300 was the problem. Swapped in a custom cable and now that thing stays up for weeks until I restart it.

Antenoralol@reddit

That happened to me until I upgraded my PSU also.

Guess it didn't like the 750W EVA unit it was connected to, had no issues since I got a Corsair HX1200i

7900 XT also.

whymeimbusysleeping@reddit

PC gaming is the best experience. As long as you don't encounter an obscure problem

tamarockstar@reddit

Those 16 pin connectors are just flat out unreliable. They should be redesigned or abandoned altogether. The 7900XT requires 2 or 3 PCIe 8-pin connectors. Your PSU comes with 4 of them. No reason not to just use those and not the adapter.

ShiftyAvatarYang@reddit

I believe you should use two separate 8 pins instead of the single 16 on that card. Glad you solved your issue!

Sremylop@reddit

This mostly depends on build quality. Corsair explicitly says that pigtails are sufficient: https://www.corsair.com/us/en/explorer/diy-builder/power-supply-units/individual-8-pin-vs-pigtail-connectors-for-gpus/?srsltid=AfmBOorgkqyKmFkB3ReLdPRRa-FPrZsbnYljcy-Dtq2C2DtU9tefvAAt