Does a CPU consume less power when it is idle?
Does a CPU consume less power when it is idle?
Any modern CPU: yes. And old 6502 at 980KHz or similar from that era probably would not. It always drew more or less the same current at the same voltage, and if it had nothing to do, it entered busy waiting. Essentially it was always busy, even if only doing this: 1. 'Do I have some work?' 2. 'no, then let's go back to point 1.`
However, the speeds you mentioned (800MHz and 4.0GHz) point to a modern setup, as does the term SpeedStep, which I mostly remember from early intel CPUs in laptops.
Work on a CPU usually follows this pattern:
- The instruction counter on a core is read, increased by one.
- An instruction is read from that place.
- that instruction is decoded (if needed) and acted upon.
- Usually, go back to the start.
This means a CPU is continually busy doing things. They are doing things mean state changes in transistors, which consumes power. Higher speeds mean more changes, thus more power used.
Now, if we could stop the whole CPU by using the HLT instruction when it has nothing to do, then it would draw no (or significantly less) power. This means that you do not gain anything from having a faster CPU do the same operations in less time.
E.g.
- Slow CPU taking 20 seconds for a job, drawing 35 Watt all the time.
- Fast CPU doing the same job in 10 seconds but needing 70 Watt during this time.
Power used (CPU wise only) would come out the same in both cases.
There is a catch though, and the faster CPU often wants a higher voltage to be able to change it states faster. That means it may draw the same current, but power used is increased.
Thus it makes sense to scale back the CPU frequency (and the voltage) when it has significant periods with no productive tasks.
To answer this part:
If I were to disable speedstepping, and let my clock run at 4.0GHz all the time; Is there a difference in power consumption when CPU cycles are spent in an application vs. in idle cycles?
Yes, it would. If clock speed is always 4.0GHz, the voltage should always be sufficient for operation at that speed. No lowered voltages, no power saved.
As for SpeedStep:
First, I heard of this was around the Pentium mobile era (P-2, P3's, Pentium mobiles CPU's, ...). Windows/Intel platforms from that era shipped with something called SpeedStep, allowing the OS the lower the speed or your CPU and to lower the voltage supplied our CPU.
These days more of this functionality is in hardware or with help from ACPI, and the CPU is not just lowered in speed, but it can be placed in one of several lower powered states (C-states). Some of these merely halt the execution of instructions, some power down part of the chip. This part is a lot more complex since powering down a whole core, flushing its cache before that, and shutting down its memory interface also takes time (and power). Ditto for bringing it back on-line. Modern schedulers do a complex dance with multiple cores, speed of cores, heat budgets, and power states. They do not do this because making a more complex chip is fun. They do it because they can temporarily increase speed (turbo boost) and save power.
Disabling all this and always running at the same speed negates these advantages. It is only sensible to do when you are going to push the chip to its limits (e.g., when overclocking) since it causes fewer power fluctuations.