Slashdot Mirror
Pushing a CPU to Heat Death, Intentionally
sdougal writes "This site is showing a Pico-ITX board running Ubuntu with no cooling whatsoever. They even let the public guess how long it would last: 'Last week thousands of you placed bets on how long the new Pico-ITX board from VIA, the VIA EPIA PX5000EG, can last without any cooling whatsoever. An ARTiGO Builder Kit was offered as the grand prize. Yesterday afternoon the voting stopped and the Naked Pico Challenge started in earnest. We simply loaded up Ubuntu 8.04, set it to work playing an mpeg-4 video and then removed the heatsink, leaving the CPU and VX700 chipset bare to the world. We recorded the event here in this video and set up a live video stream so you punters can keep a watchful eye on the PX5000EG as it works away.'"
Not true at all. Have you heard of electromigration? Its rate increases with temperature, exponentially (actually, by the Arrhenius law). Accelerated electromigration failure tests are and have been extremely common both in the industry as in research institutions.
"The agriculture ministry is not in charge of Gundam" - Japanese ministry official.
Intel had recently introduced an overheat sensor into their CPUs. They still have them, I think. There's basically a thermal probe included in the CPU packaging, and if the temperature goes over a certain critical level, the CPU starts throttling itself down, until the temperature goes down to a safer level.
Tom's Hardware (probably being paid by Intel...) did a video experiment on this. They got an Intel (early P4, IIRC) and a then-current-gen Athlon, started them both playing Quake 3, then removed the HSF.
The Intel chip promptly throttled itself down to 400MHz or so, and kept running the game (rather slowly). The Athlon crashed, hit something like 200-300 degrees C, and burned a little hole in the motherboard.
After that little stunt, AMD started building overheat sensors into their CPUs quite fast.
I saw this in action on one of my own machines, a Shuttle SN62K, a couple of years back. That machine has a known issue with the motherboard fan headers dying after about a year of use. It's also a very quiet system. I was using a 2.4GHz Celeron in it at the time. The fan header died and the fan (only fan in the machine, if you know Shuttles) stopped working. The CPU throttled itself down to 800MHz and kept right on going, for two weeks, before I actually noticed.
the formula for convective heat transfer (transfering heat from the surface of the heatsink to the air) doesn't involve thickness. A very thinly-sliced 12"x12" sheet of aluminum, uniformly heated to 200F, will transfer just about as much heat/second to the air as a thick plate of 12"x12"aluminum @200F will (there will be some differences, because the plate has a *bit* more surface area, but I digress...)
However, CONDUCTIVE heat transfer (getting the heat to go from the "hot" end of the heat sink to the tip of the fin) is directly proportional to surface area. This means that, were you to use a single sheet of aluminum foil as a "fin" on your heatsink, you would not be able to get the heat to actually travel effectively to the tip of the fin where it could be removed via convection. Thus you'd wind up with a very hot "hot" end of the heatsink (near the chip, which does you no good), and a cool "cold" end of the fin (which is worthless, as convective heat transfer is proportional to the difference between the surface temp and the air temp). If you were to instead use a thick sheet of aluminum as your "fin", that would allow the heat to easily travel from the "hot" end to the tip of the fin, where the air could take it away.
However, you can get the best of both worlds by using multiple thinly-sliced sheets of aluminum. Same cross-sectional area as the thick slice (for good conduction), and maximum surface area (for convection). Which is exactly what most heatsinks look like.