Hacker News new | ask | show | jobs
by Dylan16807 2367 days ago
The thing is, a chip generally cools just fine with heat pipes and fins up to 200 or more watts. And the built-in heat spreader gets rid of hot spots very effectively. At that power level a peltier tends to be worse than useless, even if you're putting enormous amounts of power through it. Even if you double the efficiency it's still a bad option.
1 comments
forgotmypwd123 2366 days ago
Overclocked threadrippers can hit 500W[1]. I can see peltiers and 'vapochill' style phase-change cooling making a comeback...

1. https://www.anandtech.com/show/13124/the-amd-threadripper-29...

link
hinkley 2366 days ago
I think I accept the argument that in a desktop with a giant cooler attached to the CPU, this may not improve things or make them worse.

But there are an awful lot of mobile processors with passive cooling or complex heat pipes, and it would take more convincing that such a scheme would also fail there.

I'm also wondering if it might be useful for 'race to idle' situations by extending the time until thermal throttling kicks in.

link
Dylan16807 2366 days ago
Phase change yes. Or water chillers.

I can't imagine why anyone would be more inclined to use a peltier as wattage increases. Higher wattages make that idea progressively worse unless you have some very specific and strange requirements.

link
hinkley 2366 days ago
Are you talking about the giant Peltiers that flopped in the late 90's and early 00's? Nobody is talking about that. I'm talking about micron, maybe millimeter-scale peltiers to increase the thermal conductivity of the absolute worst spots on the chip. That may mean a particular ALU, or it could mean circuits with far more layers than we can manage now (due to yields and thermal limitations)
link
Dylan16807 2366 days ago
> the absolute worst spots on the chip. That may mean a particular ALU

The worst spots aren't much worse than the median spots (over calculating silicon, not cache). Anything big enough to be a hot spot, like a big ALU, is big enough to represent a large portion of your power budget. The main goal is to get all the heat away from the chip, and putting peltiers on a large portion of the chip gives you more heat to take away. For anything significantly smaller than that, the heat bleeds out without the need of a peltier. There might be a middle ground where peltiers could make a real life difference, but I'm skeptical.

> or it could mean circuits with far more layers than we can manage now

That sounds like you're cooling the entire chip, which is the worst time to use a peltier.

link