Question Is there a way I could save my GPU from dying?

[UniformGreen]

I am pretty sure my GPU is dying.
Last week I created a post in the System category of the forums, because when I would play some games I would randomly get crashes to desktop without any errors. After I upgraded to Windows 11 the crashes turned into system freezes (that went away after going back to Windows 10 where the games would crash to desktop again). In the Event Viewer I am getting some nvidia drivers related errors. Also, once or twice I got a dxgi_error_device_hung error.

After I stress tested every PC part I would not get any crash/errors whats so ever. Prime95? Fine as butter. FurMark? Smooth with low temps. MemTest86 and the built-in Windows too? Nothing. Even the SSDs are working fine. I thought maybe it's the PSU that was at fault since it was a really old one (8-9 years or even older) so I swapped it to a brand new Seasonic 750W 80+ Gold. Issue still persisted. I upgraded my Ram that I was having since 2016 to a brand new Kingston HyperX Fury 16GB 3600Mhz. Crashing kept coming back.

Keep in mind I was going mad because I could not pin point where the crash would come from (at this point I was not getting any Nvidia Event Viewer error, just some random ACPI 2 that after some googling turned out it was from my PSU/RAM) because all temps are really low: CPU never goes above 65 degrees celsius, neither does the GPU go above 60 at full load. Also, at this point I have reinstalled Windows 10 two times and Windows 11 two times as well, including with multiple fresh GPU driver installations using DDU.

PC rig:
GPU: Nvidia GTX 1080 Ti
CPU: Intel Core i5 10600Kf non OC'ed paired with a Corsair H100i watercooler
RAM: HyperX Fury 2x8GB 3600MHz
Motherboard: Gigabyte Z490 Gaming X
PSU: Seasonic 750W 80+ Gold

After even further investigations without anything popping out I bought 3DMark and ran the Time Spy DX12 benchmark and the GPU crashed within 5 minutes! I then went into MSI Afterburner (I use it only to control the fan speed since I don't do overclocking) and downclocked the Code to -20Mhz and Memory Speeds to -50Mhz and ran it again and crashed again after 5 or so minutes. I then went back and lowered the Power limit to 90% and it crashed again. I then opened GPU-Z and I was seeing that the GPU temp would not go above 60 degrees but the hot spot would get to ~75/77 and I thought maybe it's from the memory that's going really hot? I went back into MSI Afterburner and reset all the settings and lowered the Memory clock to -250Mhz and voila, no more crashes during Time Spy.

The thing is that when gaming, sometimes it would crash after 5 minutes, sometimes after 5 hours, sometimes it would never crash so I guess Time Spy is more intensive than gaming.

Could the memory be dying or could it be some power delivery that's faulty? And if so, can I get them replaced? I know the GPU market is improving, but they are still very expensive and I don't really want to get a new once since my 1080 Ti is still a potent GPU (when it's not crashing lol). I paid $400 for it in 2019 and it's not really that justifying to upgrade to a $400 gpu now since the price/performance ratio is not really there until the prices go lower (an RTX 3070 here is ~$700 and a 3080 (non Ti) is $1000).

 

[Aeacus]

I paid $400 for it in 2019 and it's not really that justifying to upgrade to a $400 gpu now since the price/performance ratio is not really there until the prices go lower (an RTX 3070 here is ~$700 and a 3080 (non Ti) is $1000).

For almost the same amount of money, $480, you can get RTX 3060 Ti,
amazon: https://www.amazon.com/ZOTAC-Graphics-IceStorm-Advanced-ZT-A30610H-10MLHR/dp/B097YW4FW9/
which has equal performance to your current GTX 1080 Ti,
comparison: https://gpu.userbenchmark.com/Compare/Nvidia-GTX-1080-Ti-vs-Nvidia-RTX-3060-Ti/3918vs4090

And if so, can I get them replaced?

Your best bet, in "fixing" your GPU, hardware wise, is when you contact the GPU manufacturer and ask them if they can repair the GPU for you, for which you'll have to pay for + shipping. Now, GTX 1080 Ti is 5 years old GPU and GPU manufacturer may not have parts to repair it, IF they are even willing to repair it for you. But it doesn't hurt to ask, right?

Though, do note that repairing your old GPU + shipping, may cost you more than new RTX 3060 Ti. For actual price, you need to contact GPU manufacturer (AsRock, Asus, MSI, Gigabyte or whomever is the one who made it).

All-in-all, PC hardware is such where it isn't worthwhile to "fix" it. When it breaks, better to buy new one. Since repairing one, especially by DIY, requires you to know how to solder + having in-depth knowledge of that specific hardware, so you can search "spare parts" for it from the net.

 

[Eximo]

If a slight underclock solves the issues, then just leave it at that until you are ready to replace the GPU.

I agree, fixing a five year old GPU is not cost effective unless you have the expertise to do it yourself. Not much you can do about bad memory, maybe harvest chips from a parts only board, but it might be difficult to get an exact match, and they sell this chips in match sets I would presume so they might not play well together. But that means identifying the bad chip, getting a new one, having a hot air soldering station and enough experience to re-ball a memory chip. If the GPU is going bad, not much you can do.

Power delivery maybe, a lot of off the shelf components there. Still hot air soldering station and surface mount soldering experience.

 

[spentshells]

Assuming you have already ddu on multiple occasions.
Have you ever done a fresh install of windows since this started?
Not an upgrade or a repair.

You could replace the thermal pads and thermal paste. It may not solve the issue but it wont hurt anything and will have very little cost. This has worked for me in similar situations in the past.

Even recently with my rx570 where I had to back the memory off and never overclock while lowering the boost clocks. Same as you crashing to desktop without error or into a reboot.