Now that we've covered the basics of what should be required of a heatsink used for cooling today's high-end CPUs. Lets get back to the clock throttling issue but more precisely the thermal management feature as found in the Pentium 4. There's been quite a bit of discussion based on the assumption that the clock throttling feature used by the Pentium 4's thermal management is a bit 'trigger happy' and is actually affecting the CPU's performance. But what exactly is 'clock throttling' and why has it been included with the thermal management feature.
Modern CPUs are all manufactured in 0.18-micron, which means that there's an even growing number of transistors packed into every mm2
in comparison to the previous generation. Naturally that means that the denser you pack these transistors the hotter the active regions of the CPU-die will become as there's simply more heat generating transistors per mm^2. CPUs such as Intel's Pentium 4 and AMD's Athlon have millions of transistors, and the most timing sensitive circuitry is often concentrated in dense areas to maximize efficiency and cut down on any latencies. These area's such as the ALU, the FPU, but also the caches and general purpose registers are used very often and will consume the majority of power drawn by the CPU and thus dissipate the most heat and are the most likely locations for CPU-die hotspots.
Fig 5. The CPU-die of an Intel Pentium III CPU, all heat is dissipated from this small area, so a good heatsink should be used.
Any CPU, and not just Intel's Pentium 4, when in operation draws more power and thus dissipates more heat when the workload is higher. Because the power dissipation is a function of both voltage and frequency we can easily derive that a CPU's power dissipation increases linearly with frequency and with the square of the voltage (P = V2
x F). Thus a reduction in voltage will have a greater effect on the CPUs power consumption and dissipation that a reduction in clockspeed, a notion that is often used in notebooks, for example Intel's SpeedStep technology.
Halting the CPU or 'throttling' it down to for example 50% of its normal clockspeed when there’s not processing required are both similar measures taken to cut down on power consumption. Examples of which are Intel’s Speedstep, it reduces CPU frequency and voltage of the Mobile Pentium III when running on batteries, but also puts the CPU in ‘halt’ when no processing is required. AMD uses a similar approach with their PowerNow! technology as featured on the Mobile Athlon 4. PowerNow! actually changes CPU clockspeed and voltage ‘on-the-fly’ to suit the needs of the application being used, so if no processing is required the CPU’s clockspeed is reduced to the bare minimum.
Naturally there are features found on a notebook, and for a mobile CPU these are much-needed features to preserve energy and prolong the battery life. In a desktop or workstation they seem a little awkward as there’s no need to conserve energy other than for environmental reasons as they’re always plugged into the mains. So why do we need clock throttling on a desktop CPU?
As mentioned, due to the constant miniaturization there’s constantly more and more transistors being packed into every mm2
, this means that not only the heat dissipated per mm2
will increase significantly but thermal ramp rates, the rise in temperature over a given time, usually seconds, will increase significantly too. For high-end processors don’t be surprised if thermal ramp rates in excess of 50degrees/second occur for often used areas of the CPU-die. It needs no explanation that in order to counter the effects of such fast temperature rises we need more than a heatsink and a temperature controlled fan. As these would be too slow to react upon these thermal ramp rates, by the time the fan has increased its RPM to offer additional cooling the CPU-die could be well beyond a safe operating temperature.