Cooling the Surface of the Sun





This quote from ITWeek, accurately describes the impending thermal issues chipmaker's face in the near future. For this reason, air-cooling is rapidly becoming obsolete as the predominate cooling method for the modern CPU. Albeit, air, water, phase-change or LN2, the choice is yours, the goal singular; to prevent your silicon from baking. Of course that's not the worst case scenario. Most CPU's today have thermal protection diodes which prevent them from reaching critical core temperatures. Where the issue becomes much more frustrating, is in the ambiguities when attempting to trouble-shoot. High operating temps can slow a processor and even corrupt data, yet there are often just subtle hints at the true source of the anomaly.

While there are manufacturer's specifications for safe operating temps, there are also ideal temps. Where thermodynamics are concerned, a manufacturer's suggested cooling usually hovers dangerously close to the point where system errors can occur. The problem doesn't always present itself in prima facie failures such as shut-downs, or re-boots, which are easily identifiable. Heat related issues are often mistaken for software errors, where data flow might slow to a snails pace, or worse the system works overtime to correct itself. In an attempt to rectify what is in fact a hardware problem, Intel designed a "fail-safe" into their Northwood core. The P4's Thermal Clock Throttling feature has been a point of contention since its introduction. In this writer's opinion, Thermal Clock Throttling is simply an ad hoc engineering fix circumventing a wattage/heat issue. None-the-less, in the end Intel's "good intentions" won out over the astute, perhaps pretentious demands of overclocking purists.


It may surprise many to know Thermal Throttling can be active as much as 60% of the time during CPU intensive programs. It might be considered defeatist to give us the gift of Hyper Threading, only to covertly snatch it back by slowing the processor with Thermal clock Throttling. The Pentium-4 Northwood has a total of three thermal measurement diodes within. One for fail-safe, shutting down the processor in case of overheating. This diode is located deep within, and we do not have software access to it. The second diode is solely dedicated to the Thermal Clock Throttling function. Slowing the processor when certain temps are reached, and of course were not only unable to get any reading from this source, it's most often occurring without our knowledge.

Finally a diode located at the surface of the CPU, the one which your BIOS, or monitoring software reads. This diode is purposely located in coolest locale of the processor, at the IHS surface. So while your favourite voltage/temp monitoring software, albeit MBM-5, CPUCool, Asus Probe, etc. reads something like 31C, the actual core temperature is much higher. I don't want to cause a panic, however; I believe most PC-user's aren't aware of the actual temps their CPU's run at. The P4 can operate safely at temps as high as 61°C; the question is, when how often does Thermal Clock Throttling occur? How and where does it affect performance? Can it be disabled? To my knowledge, this has been attempted; perhaps successfully, regardless these questions are most often either unanswered, or unasked by the average PC-user.