Originally posted by: BonzaiDuck
Good to have your input on these matters, Dudeman, so stick around (and with 10,000 posts under your belt, I don't see you going anywhere unless you punch out from natural causes to be found with your face mashing down on your keyboard.)
"Metal layer and outermost pin-taps" -- I think this is what I was referring to -- having got it from Dr. [Somebody] in a THG article.
The reason I clarified is because you said:
there are contacts between metal leads and the silicon. At higher temperatures, these connections, over time, can separate.
which isn't the case. The outer metal layer is a metal to metal connection, not metal to silicon.
So -- on the general level -- increased resistance in metals reduces electrical efficiency, requiring more voltage? Increases in heat arise from voltage increases?
Not quite. Increased resistance in any of the materials means you need more voltage, but voltage and current have a very complex and intertwined relationship. The current is what causes the heat, not the voltage. You can think of voltage like potential energy, and current like kinetic energy. The rush of current through any material is a result of the differing potentials applied to it.
And some semiconductor materials actually display less resistance at higher temperatures and therefore better conductivity?
Resistance and heat are again a complex relationship. Light bulbs for example gain massive resistance as they heat up. When you first turn on an incandescent light, thousands of amps are surging through it. As it gets hotter, the resistance increases exponentially and it stabilizes.
MOSFETs are also bound by this thermal runaway limit. The hotter a MOSFET gets, the smaller the available channel for the electrons/holes (depending on the doping) to pass through. When I said transistors switch faster, that doesn't mean the circuit will work at a faster frequency. Unlike integrated circuit MOSFETs, Power MOSFETs are at risk of thermal runaway. Thermal noise injection and junction temperature play a complicated role in all of this.
So reductions in temperature -- even through some radical phase-change method -- might improve electrical efficiency in some components and not in others?
I'm not totally sure how to answer this, but lower temperature usually means lower resistance if that is what you are asking. The mechanisms which dictate the rate of decrease change as you hit certain boundaries (such as the Debye temperature), but the resistance is a net decrease.
See, with what you've said, and what I gleaned from the THG articles, it would seem that reducing temperatures has a questionable effect on over-clocking results, since some components improve in performance and others degrade. We see improvements of 300 Mhz or so in water-cooling versus air-cooled systems nevertheless, but this is because the over-clocker is simply boosting voltage while keeping the heat reduced enough to avoid the thermal limitations of the processor. In the long run, he is reducing the life of the CPU from electromigration through the voltage boosts.
Pretty much. Be careful thinking that higher temperatures are a good idea. I simply said the transistors can switch faster, but the circuit will operate slower. The mobility of the electrons/holes has a lot to do with this.
The electromigration effect isn't noticeable immediately, but increasing the voltage is a definite way to cause it. The higher energy the electrons are, the more atoms are ejected from the channels where contact is made. If enough atoms are removed from the channel, the transistor will become a high impedance open circuit instead of a semiconductor, meaning it is always off.
So people reduce temperatures through water-cooling, then pump up their VCOREs way too far beyond the manufacturer spec -- assuming that the resulting temperature regime means they're not going to damage anything. But -- they do anyway -- for the voltage increase.
Right. People are way too worried about temperature. Running at a higher voltage is far more detrimental. The Q6600 is cooled very much within spec using the stock heatsink. There is no need for water blocks just to run it at stock speed, which I have heard many people say.
Here's an observation about the ThermalRight Ultra-120 (and Extreme) cooler, and other coolers built with that design. SVC told me they shave off less than a millimeter of heatsink base to flatten it. That's really a thin sliver of copper. If one were worried about the reduced pressure on the heatspreader, you could simply add (nylon or metal) washers to the bottom of the springs in the mounting assembly to compensate.
The extra force would be deflected onto the frame instead of a relatively small area above the die. I understand what you are saying, and it would probably add some marginal increase in pressure, but nothing like what the convexity of the bottom would have added.
Of course, you're saying that this pressure is concentrated along the ridge of the convex surface of the heatsink base, and that removing that ridge sets up an entirely different "pressure regime." But if that were so, wouldn't TR recommend against mounting the cooler in one direction as opposed to another?
The convexity of the bottom is supposed to be circular, meaning that it can be mounted in any configuration as it should be uniform.