Does anyone know what the performance difference is between really cool systems and really hot ones(by really hot, I mean stable at high temps). As in, how the temp levels affect the items. The specific reason I ask is that as temps increase or decrease the volume of materials increase or decrease(sometimes this is in a non-direct/linear range as with water/ice). I want to know(strictly for curiosity's sake) if the reason overclocking and overheating causes crashes is because the material involved(metals, plastics, silicon, etc...) expands and then operates improperly. An example of this being like touching two connections together where they shouldn't, or the connections being broken apart due to the expansion. Basically, it's a question about the engineering aspect of the stuff involved. Why does overheating cause crashing? I also know that electrical current flows much better at lower temperatures. Is the opposite true on the reverse? Hotter electrical parts operate improperly? I would think this is related to the activity of the electrons in the metal interfering with current flow or diffusing the current's flow in an outward fashion instead of in a circuit flow. I know this is a lot of stuff that someone else might already know, but it's kinda hard to put it in simple terms of search engine to find it, so I ask here. Can anyone give any input on this to shed some light on the matter? I'll be looking around the net for info, if I can manage to find any leads. Thanks and I hope I'm not sounding too damn dumb, here. (I'm the kind of guy who asks, "why does it work?" not just "does it work?" if that helps put my questioning manner in a more understandable light.)
Overheating causes crashes due to how heat affects electricity flow. At higher heat, it takes more current to get the same voltage (which is why superconductors tend to be in labs at -200c), which in turn fries/damages/whatever the chip - silicon melts at something like 1200c. In theory, you could run your chip at some insane temperature, but as the temperature rises, the power draw does, which in turn produces even more waste heat, leading to a loop of destruction. The power-down is more of a safety feature. When the heat is kept in check, this isn't an issue. Might not be dead-on, but that's the general idea. However, I'm pretty sure you could stick your processor in the oven overnight at the hottest temperature, and as long as you don't melt the gold from the contacts, it should run fine afterwards. Don't try it, but the heat is only problematic when electricity is flowing. Expansion is pretty much negligable, especially over the temperature range at which computers typically operate. Some people make that arguement, but silicon doesn't undergo the same roughly ten percent change that water does when frozen. This is more guesswork on my part, but I've yet to see definate evidence that thermal expansion isn't negligable over the usable lifetime of a microprocessor (say five years to be generous, depends on what kind of user you are). My grandparents have some superold Mac, probably from 1994 or so, and it still runs like it did the day they bought it (the lack of internet connection helps that quite a bit!), and it's probably been turned on and off thousands of times. In fact, I'm surprised the hard drive still works, but there you go. Oh yeah, posting this in electronics or hardware might have been just a bit more relavent
Very high temperatures can permanently deform circuitry. About the properties which create instability without permament deformation... Unless someone knows for a fact otherwise, I wouldn't rule out thermal expansion from being *one* of the factors that can cause instability.
Cool... Cool responses, thanks. Kinda confirms a little of what I thought, putting it in a slightly more understandable format. Sorry for posting in the wrong area, just couldn't figure where to put it... :\ Back to looking for the next inane question or idea, hehe.
I have just been sitting here thinking for a bit about the whole thermal expansion thing and have come to the conclusion i don't know enough about what the cores are actually made of to comment. What actually makes up a core? is it just silicon or are there tracks "printed" onto it like a circuit board? The thing to remember about thermal expansion is that it's all relative, nion all materials expand to some extent when heated and so it depends on the differing rates of expansion between the different materials. It's certainly something to think about. A question thats always interested me is why some CPU's overclock better than others, not just in general, but specific steppings etc. Surely if they are all made on the same machine out of the same materials then they should perform within a pretty tight window.
Actually, I doubt that they turn a setting and pump out 2.8 GHz, and then set it for 2.6GHz, etc. I believe that maunfacturers label CPUs according to the stable clockspeeds that are measured. So even if two chips are stable at the same clockspeed, they have different tolerances when it comes to increasing the speed the chip was set to. This is experienced with ram as well.
Probably worth making a seperate thread for this, but as a REALLY brief answer, it depends on the "quality" of the silicon the chip was produced from. ie, dies taken from the center of a wafer will be of much higher quality to those taken from the outside. Thats kinda the reason batch numbers are important when choosing a good clocking chip. Steppings just tell you which FAB made the chip, on which machine, and at what time. (Along with default voltages etc...)
The thermal expansion rate of silicon is on the order of a few micrometers per meter per degree C (that's like 0.0001%). So if you heat your processor up from 20 deg C, to 80 deg C, you only see an expansion of maybe 0.01% of just the silicon, which is pretty negligible. However, the other materials that reside with and near the silicon, that's a different story. Cheap solder will melt at 90 something deg C, so you can start shorting things, soldering components together, and other bad things. The fiberglass PC board will also smoke, burn and warp at around 100 C. So, I don't think thermal expansion has an effect on system stability when overclocking. It's more resistance, high frequency, voltage supply/demand related than mechanical expansion related. -L6
As Firehed explained, you can get thermal runaway with high temperatures; the increased resistance means increased heat output from the chip and, if it's not adequately cooled, you can burn it out fairly quickly. However, the main factor in temperature affecting chip lifespan is that of the leakage current. In all semiconductor devices, increasing the temperature of the FETs (transistors) increases the gate leakage current (the current escaping through the gate oxide), because the electrons have increased kinetic energy and so more can tunnel through the gate. If enough electrons have sufficient energy to travel through the gate, the voltage across the FET will drop and it'll switch to a 0 when it's meant to be a 1. That happening in multiple locations across the chip will cause crashes and instability; long-term repercussions can include electrons "eroding" the oxide, rendering the transistor useless. Interestingly, the leakage at high temperatures also means that hot chips won't be able to hit clock speeds as high as cold ones, because the voltage takes longer to ramp up (more info) if electrons are escaping through the gate (which is in part why phase change coolers work). Edit: 3000th post
Not so sure on that, eg, for a typical PIC, max operating temperature 85C, max storage temperature 150C. Get too warm for long enough and those carefully dosed atoms will be diffusing across the N-P boundaries giving a permanent change. (I think that's why some LEDs go dimmer over time at high current.) The old MFM/RLE hard-drives sometimes had problems due to simple thermal expansion. I came across one PC where the only fan (in the PSU) had failed and the system always crashed after half-an-hour or so. Drive was far too hot to touch, the tracks were no longer in the same spot as the head expected them to be, so file read errors. Modern drives don't work that way, but they still don't like too much heat. And in the days when ordinary DIL sockets were used for chips, one tip when trouble-shooting was to try pressing chips back into the sockets, as they could walk their way loose with repeated thermal expansion & contraction.
In reference to the above quote; cool information. I'd never thought of the whole chip coming out due to repeated heating/cooling. Kinda funny, when you think about it. A sort of elapsed-time chip dance out of the socket.
