Hacker News Reader

Comment by pram | Hacker News Reader

I’m not sure I’d want a card that’s been at full utilization, 24 hours a day, for who knows how long.

On the other hand, you can be sure these cards have made it well past the "infant mortality" part of the bathtub curve.

In fact, military and other high-reliability applications specify "burn-in" periods where components are operated at elevated temperatures for several hundred hours --- they are effectively "used":

http://reliabilityanalytics.com/reliability_engineering_libr...

I recently learned a really fascinating stuff about semiconductors: as they heat up, the doping atoms tend to move, by diffusion. Heat = more diffusion.

Yep: atoms in general move more with more heat. In solids, the energy required for movement is higher, but it still happens. The more heat available, the more energy you have to break bonds, the more diffusion (movement) you have.

I have a feeling that's not the main reason for heat failure of semiconductors though. (Not to put your comment down, it's an interesting fact and I'm glad you mentioned it.)

The weak link on a lot of hardware is the capacitors - I'm not sure about GPU cards, but if that's the case then you sure won't be replacing that surface mount stuff when it fails.

Why? All you need is a heat gun and some spare parts.

Another fun fact: Chips are designed to last years at full power and max temperature.

It's actually the temperature cycling that's far worse for them.

Plenty of available reading on cryptocurrency mining forums.

I've read that thermal cycles do more damage than constant running.

Yep, this applies to most things. E.g Engines.

If I can buy a 1080 for $100 when they flood ebay, I'm willing to take my chances.

Reply on news.ycombinator.com