Hacker News Reader

Comment by userbinator | Hacker News Reader

They have throttling, and the newer ones are definitely "thermally governed", but without a heatsink to provide thermal mass, I'm not sure if they can react quickly enough to stop going over the limits. That said, a CPU with a heatspreader will have more mass than a bare die.

View on HN

I repaired a piece of manufacturing equipment in 2018. It was terribly slow, even more so than one would expect from a first gen core i3. The heatsink had spent 8 years, hovering a few millimeters above the cpu. The Intel thermal paste was untouched.

I still remember a GeForce MX440 I bought once. It was stuttering as hell in games. Overclocked it for better performance. Still stuttering.

Looked at it and two of the plastic clips (yeap, no screws back then lol) holding the heatsink+fan were snapped so the heatsink wasn't even touching the die... WTF!

That was pretty hilarious. It worked fine for years after I fixed it, long enough for caps to blow and be replaced.

Basically, they're fucking sturdy.

To add on your anecdata: I've seen an MX440 with a fan glued to the die. Without a heatsink and I don't think that was a thermal glue. It worked badly, but it didn't die.

What is "quickly enough"?

Isn't the reaction time today somewhere on the order of a few milliseconds? How much could a component overheat in that time? The maximum current, though high, is still limited, after all.

Also, there is "thermally governed", and there is that hard reset once a critical temperature has been reached. I'm admittedly ignorant, but a "extremely short reaction time" with regards to a critical event seems like something that one could reasonably expect.

"I static'ed my component X" is something one hears frequently, but I don't think I've ever read about anyone frying their CPU anytime in the last decade.

Edit: Looking at a somewhat recent paper [1] studying CPU frequency transition latency, even the highest latencies were still within 70µs(!). I think one could reasonably expect a similar or better latency regarding critical protection circuitry.

[1] https://www.cse.iitd.ac.in/~sbansal/col862-lpc/ftalat.pdf

My experience is mostly with Intel mobile CPUs, which are mounted as bare die without a heatspreader, and those can jump 20-30C a second even in normal operation with a heatsink attached.

Sure, but that "second" you mention is an eternity compared to the max 30µs (see referenced paper) it takes a CPU to switch from highest to lowest frequency.

Why would you expect critical temperature threshold shutdown latency to take any longer?

Reply on news.ycombinator.com