[Retric]

You can’t tell from an eBay listing that a given card was undervolt or overclocked. On average it’s far more risky than buying a gamers card which, so their best avoided.

Also, undervolting isn’t always the correct choice it depends on how valuable the coin being minded is relative to energy costs. Someone mining in their dorm room for example may not be paying based on electricity useage.

[Sholmesy]

Citation needed?

My understanding was that gaming cards are pushed far harder, at higher temps, with fluctuating power and thermals, which causes more issues than a single stable power limit and temperature

[dahart]

Where is your understanding coming from? There is no such thing as pushing cards “far harder, at higher temps” than when mining or doing other compute tasks that run the GPU at 100%. Failure rates land squarely on the side of higher temps, and the reasons are well understood https://electronics.stackexchange.com/questions/444474/can-i...

You might be thinking of spinning disk drives rather than GPUs. A lot of people suggest that leaving an HDD powered up and spinning is better than spinning up and down frequently, due to the temperature going up and down a lot and the added wear on this mechanical device. This is completely different from a GPU though.

[mlyle]

Higher temps are bad, but thermal cycles are equally bad or worse. Different things on the card have different thermal coefficients of expansion. Getting warm and cooling makes everything flex and stresses solder joints, wire bonds, and thermal interfaces.

Miner cards have longer, sustained high temps. This is bad for life.

Gamer cards have lots of thermal cycles. This is very bad for life.

Miner cards are more likely to be undervolted to improve power efficiency and thermals. This is good for life. (Lower peak temperature, less electromigration).

Gamer cards are more likely to be overvolted and overclocked to improve peak performance. This is very bad for life. (higher peak temperatures, more electromigration).

https://www.dfrsolutions.com/hubfs/Resources/services/Temper...

[Retric]

That’s testing for thermal cycling over wide temperature ranges or longer lifespans. GPU’s are used indoors and don’t have a very long lifespan.

The major risk factor for GPU’s is electromigration which is a major factor in GPU lifespan and directly relates to usage. A 40 hour a week gamer is extremely rare, but a mining GPU is pulling 168 hours a week.

[mlyle]

Electromigration is a small risk factor in any kind of reasonable life. Especially if not overvolted (which is something that mostly gamers do-- miners are more likely to undervolt).

Solder fatigue breaking of solder balls is common. I have fixed lots of GPUs by reflowing them. GPUs do cycle over a large temperature range-- delta-T can be 50C+. While maps are loading, etc, you can have delta-T's of 25C+ every few minutes.

Indeed, you have lots of people doing this:

https://turbofuture.com/computers/How-to-Fix-a-Dead-Graphics... https://www.instructables.com/How-to-repair-your-Graphics-Ca... https://www.ifixit.com/Guide/Temporarily+Repair+a+Lost+Cause...

This is a thermal cycling induced failure mode. (Of course, a home oven doesn't accomplish proper reflow, so this is more of a "fix things for a couple months" trick as described in the posts).

[Retric]

As all of those links mention it’s useful for a minority of dead cards as most are failing from other causes.

That said, specific manufacturers can always introduce defects so your mileage may vary.

[cmeacham98]

Mining cards are run near always, while gaming cards are usually pushed only a small number of hours per day.

I have no data to back this up but anecdotally this makes a huge difference in wear.

[pantalaimon]

GPUs are not mechanical parts (well save for the fans but those can be replaced). I would imagine thermal stress from heating and cooling would be the biggest issue - you don't get that under constant load.

[iforgotpassword]

Heat = bad for silicon. Also electromigration. And probably a couple other effects I don't know about.

[R0b0t1]

Heat is bad, but MTTF at usually achievable temperatures is hundreds or thousands of years.

Electromigration matters on the order of 100 years.

[namibj]

Well, tell that to my chipset's SATA controllers (Sandbridge generation).

Intel messed up their life expectancy calculations, and thus they died after 2~3 years. I think 2 are still alive.

[locuscoeruleus]

Huge difference in wear, yes. But not in the direction you think, I think. Warming up an cooling down is more damaging for a card than running constant temperature. It 'jiggles' parts more.

[gsich]

With ETH you underclock core and overclock memory. Core will lead to vastly lower power consumption, while still maintaining hashrate.