I understand the basics of how and why speedstep does what it does but im curious...
How exactly does it work?
What is the criteria used to calculate whether the clock speed should be increased or dropped?
Does this mean that when clocked down its truely running at that speed or is it kinda clocked down in a way that once you start putting load onto the system that it instantly reacts increasing speed to compensate?
Is there some kind of delay when increasing the clock speed which can lead to an initial bog down of the system until the cpu catches up?
Im kinda looking for the technical aspect of this and of course answers to the questions above. Hopefully, you guys kinda understand what im asking if not let me know and ill specify. JumpingJack i know you probably have some input!! lol
It is quite simple really, power saving technologies usually revolve around changing the clock/voltage of the processor.
Speedstep (and the AMD equivalent Cool 'n Quiet or PowerNow!) operate by assessing the load request on the CPU, if the CPU load is low or idle it will enter a different P state (a standard specified in the ACPI specs). What that does is to actually change the clock multiplier down a notch or two, thus the processor is actually clocking out slower at idle or low utilization. Similarly, Vcore is dropped down automatically, in both cases lower frequency and lower voltage translates into lower power consumed, less heat genarated and less power dissipated.
At this point, I don't think i have tuaght or communicated anything new to you. Honestly, you could go look at the specs and find out what criteria is needed for the different power states.
Also, to answer your other question, on the time scale of the transistion it is basically instantaneous. It happens over several cycles, but in real time this is on the order of microseconds.
Intel is usine what is called a C1E state in C2D, which AMD does not have, it is a processor specific state (as opposed to the P states), which yields higher power saving benefits.
Neither of these save power at full load though, since at full load voltage and power will ramp back to the stock/settings as specified by the processor/bios.
C2D has a unique power saving mechanism in which it can power gate certain parts or blocks of the core that are not being used, it can do this even while running at full load. This is particularly novel, but does not fall within the ACPI spec for power gating, it is likely Intel has several patents on this so if AMD is to follow suit, they will need to either license or figure a different way outside of the patents.
More to add:
For AMD's CnQ, the idle state will be 800MHz (Sempron, S754 A64) / 1000MHz (Other A64 / A64 x2) and teh voltage will be 1.1V
For Intel's EIST, the idle state will be x6 multiplier (C2D desktop) / x14 multiplier (Netburst desktop).