What kills gpus?

[RAYSLA]

I heard that they usually die because of things like temperature cycling or high temperatures but i have seen people with 900 or 10 series that have their gpu's at 86c for extended times and they keep running with no problem while other gpus like the 20 series with better cooling solutions randomly die.

is there any other way to extend the life of my gpu?

 

[Shrek]

Set the power maximum to less than 100%

 

[freeagent]

I run the shit out of mine. Thousands of hours in game, also even more thousands of hours running F@H. I run at full power limits, and as fast as the memory and core will run stable as a table. I have always run them that way.. only had an XFX die that was not by my hand and warranty fixed that.

I don't turn off my computer unless I am digging around or changing something, or cleaning.. I work them, not baby them. That is a lie.. I baby them by keeping them cool and clean lol..

 

[Frick]

Heat, power, manufacturing.

 

[Konomi]

Humans.

In all seriousness, typical failures are caused from various forms of degradation. If you're concerned on a new purchase - drop the power target, don't overclock, regular thermal paste changing and generally keep it as free of dust as possible. Also cap framerates in games to your monitor's refresh rate (or less), as that'll help reduce the workload on it.

 

[R-T-B]

Humans.

And frogs.

 

[Sithaer]

As mentioned before, various reasons and some cards simply have delayed manufacturing defects that will only start showing up later.
I've had my old 8800 GT die on me after 2.5 years like one of the memory modules kicked the bucket one day and then the same happened to my bro's 8800 GTX and to my friend's 8800 GTS and then another friend's 9800 GT cause it was a common issue with those cards.
I've also had a 560 Ti die on me cause that card simply fried itself to death in a year after I've bought it, I was new to all of this stuff back then so I couldn't help it. 'Most likely all it needed is a good ol repasting..'

Then there is my bro's GTX 970 that is still kicking and alive and yet he never took care of it nor repasted it so that is one heck of a survivor. 'To be fair its a Gigabyte Xtreme 970 and those were overbuilt'

Personally I'm almost too careful with my GPUs nowadays like I always undervolt them and cap my FPS and whatnot and so far all of my cards survived their years of service ever since that 560 Ti. 'knocks on wood'
My current 5070 has 2 profiles saved in Afterburner, one for the maximum stable OC+Undervolt and another is stock with undervolt which is very efficient I must say. 'I switch between those 2 depending on what I'm playing, like if I don't need the extra performance then I just run the stock one'

And frogs.

This might be the biggest culprit here.

 

[Dr. Dro]

Baking them on a crufty, cramped case or pumping them full of current non stop without maintenance or cleaning for years on end mining crypto will usually kill a GPU, yes. Thermal cycling? They're designed to handle that, not that it's a consideration when your load temperatures range from ambient to high double digits anyway. PCBs are copper, after all.

If you keep your GPU operating in a good temperature range, allow it plenty of airflow and do regular maintenance, it will last a lifetime.

 

[Ruru]

Pretty good points mentioned already. I'd say that complexity and the build quality are also culprits of a possible failure.

 

[Macro Device]

They mostly die from factory defects (some RTX 2000 non-Super cards, for example, had defective VRAM from Hynix and died randomly), bad quality power delivery (be it just as simple as a pisspoor PSU or the circuits at home being a whack) and human errors. Operating within reasonable temperatures (<90C for the die and VRAM chips; <95C for VRM) can last virtually forever; at least most cards become super obsolete before they physically die.

Don't make a big deal of it, check if your PSU and overall electricity situation is fine, keep temperatures sane and don't overvolt your GPU and you'll be fine. And if it dies, it's a factory defect.

 

[Sithaer]

They mostly die from factory defects (some RTX 2000 non-Super cards, for example, had defective VRAM from Hynix and died randomly),

Apparently my Asus TuF 3060 Ti LHR also had the Vram of doom with the Hynix x005 number on it that supposedly fails over time. 'Checked it when I was repasting the card'
I've bought it second had 2 and half years ago and I've tried to babysit that card as much as I could and it did not die on me and now its sold off to a system builder who checked it and told me that its all good so I was paid. 'oh well its not my problem anymore huh..'

 

[mclaren85]

Most definitely too much heat-cool cycles. It will eventually cause solder crackings for sure

 

[ShrimpBrime]

Under normal usage, a video card should last as long as the capacitors. Most solid state caps are 100,000hrs.

Why do some just die? Cause complex circuits may not be as easy to design based on parts pricing and availability or it was just a design flaw somewhere.

Other than that, humans have working electronics flying through space thats older than most of us.

 

[Shrek]

I thought it was fans that went first.

Which leads to the thought that a well-ventilated case will help.

 

[ShrimpBrime]

I thought it was fans that went first.

A quick look at a random seasonic power supply

G12 GC (2024)

Seasonic G12 GC PSU designed for reliable performance, boasts gold efficiency rating, ensuring optimal energy use and reduced operational costs.

seasonic.com

has

• Fan life expectancy 25,000 hrs
• MTBF 100,000 hrs

Yeah, thats true. Fans. But a lot of cards come with silent mode too. So that saves the fan usage.

 

[freeagent]

Silent mode is standard on my card. Only 1 bios.. and I have to use their software lol..

Edit:

Just send it man, don't be shy

 

[rusty caterpillar]

Other than that, humans have working electronics flying through space thats older than most of us.

Those technologies costs are billions, you not gonna see the missing ROPS and FSR4 flying around in space they'll BSOD at the launch... of the shuttle.

 

[Makalite]

Using them.

 

[ShrimpBrime]

Those technologies costs are billions, you not gonna see the missing ROPS and FSR4 flying around in space they'll BSOD at the launch... of the shuttle.

Indeed, ask Musk. Blew up on the launch pad. Probably a Tesla card missing rops XD

 

[rusty caterpillar]

Indeed, ask Musk. Blew up on the launch pad. Probably a Tesla card missing rops XD

No, no he used Intel GPU with microcode issues, experts said "even if it didn't blew up at the launch the impurities in the silicone would killed the core later in the middle of the mission, crash was imminent" XD

I heard that they usually die because of things like temperature cycling or high temperatures but i have seen people with 900 or 10 series that have their gpu's at 86c for extended times and they keep running with no problem while other gpus like the 20 series with better cooling solutions randomly die.

is there any other way to extend the life of my gpu?

Heat is the biggest killer IMO. Same as for any electronics, degrading maybe first and than dead.
Cap FPS to reduce power = less heat.

Rear fan of the case should work at least 50% of his capacity or more at all times to exhaust heat generated by the GPU including the fact that will enhance airflow of the backplate of the GPU.

IMO VRAM(memory junction) temps killed most of NV 2000 series, Micron memory modules, same for NV 3000 series.
Prioritize cooling of the VRAM and VRM over GPU core.
Try to get GPUs that have exhaust NOT obstructed by plastics and logos as much as possible.
https://www.techpowerup.com/forums/...odel-in-terms-of-quality.335281/#post-5537880

Another reason for dead GPUs : Cracked PCB during installations or shipments of the GPU as the PCB for this series was thinner than older versions.

Lack of lead in the soldering. Nothing you can do about this one except: don't twist your card during installation and use anti sagging support.
Some of the manufacturers returned to lead soldering as example : PNY. I've seen on their website for a 4000 series card the warning : "Attention contains lead"

This thread has enough info have a good read of it: https://www.techpowerup.com/forums/...-temperature-regular-value-or-hotspot.334188/

I run the shit out of mine. Thousands of hours in game, also even more thousands of hours running F@H. I run at full power limits, and as fast as the memory and core will run stable as a table. I have always run them that way.. only had an XFX die that was not by my hand and warranty fixed that.

I don't turn off my computer unless I am digging around or changing something, or cleaning.. I work them, not baby them. That is a lie.. I baby them by keeping them cool and clean lol..

You have a pretty good case, so a solid base for good airflow around GPU.
The reality IMO is that most people will look at RGB and looks when they choose their prebuilt or case to build a PC and forget the rest, while a good amount of ppl look at the price of the case.
And just to detail the issue further: in both situations you not have good airflow for the GPU, or the CPU has cooling priority over GPU. And the later is an common mistake, they just cool well the CPU that consume roughly around 100 Watt and ignore the GPU that consume 200-300 Watt, basically GPU backplate airflow is suffocated by the big CPU air coolers or the CPU has AIO with 3 fans running all the time while GPU fans are stopped, I can't get more clear than that

 

[CheteriagalenTriagalnik]

I thought it was fans that went first.

Which leads to the thought that a well-ventilated case will help.

Yeah, but this is a relatively easy fix. Even after warranty ends, you can usually find spares on sites like AliExpress. Failing that, you can deshroud the GPU and put a few good-quality case fans on top.

In any case, as with what people have said here - heat kills over time. So, undervolting, underclocking and power limiting the card helps. The lower you can keep the temps, the longer it -should- last. I say should, because you never really know. GPUs have plenty of components that can go bad. Besides the die, there are also VRAM chips, VRMs and various other components. Lowering the power consumption and improving the cooling helps, as does regular maintenance (usually just air dusting, repasting really only needs to be done if original compound gets too dry), but ultimately you can't ever be sure. Maybe helping the card out against sag with a little brace could also protect it from PCB damage over time, but this is really only a problem in extreme cases.

Presumably, manufacturers that offer longer warranties might be a safer option for buying a new GPU, as the longer warranty could imply they have went with higher-quality components, but again, you simply don't know the longevity until it's too late.

I have a Gigabyte Xtreme Gaming GTX 1080 that's 9 years old and still kicking, so you can definitely get a lot of use out of a GPU if you really want to.

 

[lexluthermiester]

I heard that they usually die because of things like temperature cycling or high temperatures but i have seen people with 900 or 10 series that have their gpu's at 86c for extended times and they keep running with no problem while other gpus like the 20 series with better cooling solutions randomly die.

is there any other way to extend the life of my gpu?

Heat kills electronics. Find a way to keep your equipment cool, sub 70C, and they will last you a very long time.

 

[ir_cow]

Humans with stupid software kills GPUs.

 

[:D:D]

Stupid software? Care to elaborate?

 

[lexluthermiester]

Stupid software? Care to elaborate?

Daft dimwits running Furmark for 24hour burn-in sessions comes to mind.. Totally wasteful and has the potential to kill GPU hardware. 15 to 20 minutes is all Furmark needs to test and verify stability.