Does undervolting and underclocking reduce power usage and heat output?

[Generic User #2]

I've noticed that I hit or go above 1.6 GHz usage on my Sandy Bridge CPU - I calculated this by looking at CPU Usage % * current clock speed (according to CoreTemp).

I'm just wondering, if I set my max clocks to 1.6 GHZ and I end up using 50% CPU time, is that equivalent to a 3.2 GHz CPU running at 25% CPU usage? assuming that the faster cpu is running at a higher voltage.

or another way of looking at it, does a CPU use more voltage as its cpu time usage increase? or only when it needs to bump its clockspeed?


this question is actually mainly for laptops....but i trust the guys over here more than the laptop forums.

EDIT: I actually care more about power usage than heat output...but getting benefits on both would be good(I believe they're tied together anyways).

[enaberif]

Intel cpus have power saving fumctions built in which undervolts the processor by default.

[Generic User #2]

ah sorry, i forgot to stress one point:

i have seen the cpu clock all the way to full speed(and increase its voltage). after 'doing the math,' i realized that it was running at 1.6ghz of the max 4 ghz*. with that realization, if i set the max cpu speed to 1.6 ghz, that should stop the voltage from increasing, which should use less power. I'm just wondering if that makes sense in real world usage.

*i'm assuming that 'Ghz used' = CPU Time Used % * Current CPU Frequency

[frontier204]

Yes underclocking / undervolting reduces heat and power consumption (e.g. I've cut ~35% power consumption and 20°C load temps on my Folding@Home + HTPC, but that's a better-than-average result because it has a very inefficient Phenom 9950 CPU). If you can undervolt without underclocking you will save on energy AND power. (e.g. lower your CPU's voltage but don't touch any clock speeds).
However, unless you load your CPU at 100% at all times don't undervolt and underclock at the same time. CPUs use a LOT less power when they're idle, so the faster they get stuff done, the more time they can switch to idle, saving you energy in the long run. For a general purpose computer, you're better off at default than underclocked because stuff takes longer to do.

Originally Posted by Generic User #2

I'm just wondering, if I set my max clocks to 1.6 GHZ and I end up using 50% CPU time, is that equivalent to a 3.2 GHz CPU running at 25% CPU usage?

Nope - actually the 3.2 GHz will draw quite a bit more power even at less usage. The equation for dynamic power for CMOS circuits (like your CPU) is: P= αC(V^2)f where f is the frequency and V is the voltage.
There's that all-important squared term for voltage which is different if your CPU is operating at 1.6 or 3.2 GHz.
EDIT: Note this is a theoretical equation from a textbook - you also have to consider static power (e.g. the power your components take no matter what they're doing).

[Generic User #2]

ah, frontier204, maybe i should rephrase that question:

does a CPU at 1.6GHz with 50% CPU usage running a video encoding use as much power and finish as quickly as a CPU at 3.2GHz with 25% CPU usage?

The key points being: a SPECIFIC amount of work is being done and CPU usage is NEVER at 100%

The main reason I am curious is why does the processor not wait until it hits like 90% CPU usage at its current clockspeed THEN bump the clockspeed up to the next setting. During my testing, no product of CPU clockspeed and CPU time usage has ever produced a number even close to 1.6GHz


I already know that it will use less voltage. It seems like the scenario should produce identical results, EXCEPT for the voltage increase in the latter case; therefore, logically, I should stop the CPU from going above 1.6GHz(I can reliably say I will never need more CPU power than that - I've already run my most strenuous tests on it) and then undervolt from there.

[frontier204]

How are you getting the video encode to only use 25% of the CPU? Did you intentionally bottleneck it with a slow hard drive + no compression or something?
The same CPU at 3.2GHz should still use more power - at least that's what I found in practice. When I did a little testing, there wasn't much difference in power consumption at the same frequency between partial load (e.g. 30%) and full load (100%). The amount of time it takes to finish really depends on other stuff like how the software is programmed and what you're doing to bottleneck the CPU to somehow make the task only take up the 25% CPU...

The CPU makes its decision to raise or lower clock speed in a period of microseconds, so you're not going to be able to "view" its decision-making process without some professional debugging hardware attached to your motherboard (i.e. you have to work for Intel or AMD for such hardware). For all you know, your CPU did hit that midpoint frequency but only for a very short period of time that your monitoring software couldn't pick up.

If you really want to dive into optimizing power consumption you will want a way to measure this because there are a lot of things that come into play, like how much power your computer draws while idling. If you are using a laptop or UPS, then the easiest course of action is to charge the battery to 100% and run both cases to see where the battery gets to. For a desktop just hooked to the wall, get a power meter (e.g. Kill-A-Watt or Blue Planet).

For example on my old home server (Athlon X2 5200+), I had something along the lines of:

• Undervolt only (2.7 GHz): 110W
• Underclock + undervolt (2.2 GHz): 95W
• Idle: 60W

There was a difference between not underclocked and underclocked, but the idle was still low enough that I'd prefer to finish the task earlier and idle rather than sit at underclocked but at load for a while.

This can change quite a bit depending on the hardware. For example my other Folding rig had something like this:

• Power consumption at 2.8 GHz: 235W
• Power consumption at 2.2 GHz: 130W
• Power consumption at idle: 85W

Note that was a Phenom 9950 with that 2.8GHz setting being an overclock. (I used the lowest stable voltage for both the 2.8 and the 2.2 GHz.)
If your computer's like this it may be a toss-up on which route you want to take, assuming you don't care about the task finishing early. This case is more rare (unless you know you have a horribly inefficient CPU), so you're more likely to fall into the first case where finishing early is the best way to save energy. If power and not energy is your only concern (e.g. consuming too much power => very loud fans) then of course underclock until you're satisfied. However, if you actually do that, you would've been better off buying a lower-end CPU to begin with, as the "underclocking" and voltage optimization is done for you, and it costs less money.
EDIT: I'm only doing an underclock because I turned my old gaming PC into a HTPC+server and I don't want to sell my old hardware.

[Generic User #2]

sorry, the video encoding was just an example - i didn't actually run that test. i just needed an example of a finite workload.

EDIT: alright, thanks for all the discussion :) .

I was looking to get a laptop; specifically, the Sony Vaio Z21. After observing my desktop CPU usage, I realized that even the lowest-end CPU (2.4 GHz) would be 'too fast' for my needs. Therefore, if there was a way for me to turn the performance into battery-life, all the better for me.

[frontier204]

I own a Samsung QX411 with a Core i5 2410m (100 MHz less than what the Z21 asks for) and the CPU draws SO little at idle that the best thing you can do is finish your tasks early to make use of as much of that idle mode as possible. I can't measure this laptop with a power meter because the battery isn't removable and would mess with the readings.
Windows usually reports 6h battery life for that laptop at idle but something like 2-3h at load, although the most I've ever tried was 3.5h (I find a plug long before needing to drain the battery, working in a university and all).

On my laptop's i5 using OCCT + windows to limit CPU I get:
0.79V @ IDLE (that's why you want to finish tasks early)
0.91V @ 1.2 GHz
0.94V @ 1.6 GHz
0.99V @ 1.8 GHz
1.05V @ 2.3GHz Turbo off
1.15V @ 2.7 GHz/Turbo) .

I used to use RMClock for playing with my laptop's voltages, but it doesn't play well with 64-bit Windows due to unsigned drivers: RMClock Utility. Products. CPU Rightmark

There could be some merit for turning Turbo Core off but going farther than that won't get you much. (One tick of screen brightness less will save more on the battery.) As enaberif said, the laptop will automatically choose one of the lower states if it doesn't need all the CPU, so aside from turning turbo off you don't need to cripple your own CPU.
I do notice the Z21 is a relatively expensive laptop ($1950?) - I picked up my QX411 for $1000 and it has a dedicated GPU which this thing doesn't have internally...
Why not get a cheaper laptop with less CPU? It sounds like you should be looking at an i3 laptop if your usage is that basic. (One thing my gamer friends found out for me in buying $2500-3000 laptops is that there's no such thing as "future-proofing" them.)