I would like to know how to determine how much power a processor usage. I've done so much googling, but I have not recieved a single answer that is useful to me.
I have a e6300 running @ 2.8Ghz, stock voltage (1.288 after vDroop), and I would like to know if there is some formula, or a method to isolate the power consumption of the processor, since it is one of the most power hungry parts of a computer. I like to undervolt things while also maintaining an overclock, because I like to lower power consumption while increasing performance.
I would also like to know how power usage scales with voltage. I have been wracking my brain for a good long time trying to find an answer to both issues, so being able to put them aside once and for all would be a blessing.
Determining power usage of a CPU & scaling with voltage
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
1. Some sites list them, can't tell you exactly because I read Russian hardware sites 
. At about 3 GHz it shouldn't be more than 60 W though.
2. AFAIK, it's rather complicated but as a rule of thumb it's a square relationship. So over/undervolting has a much greater impact than over/underclocking.
Edit: found some formulas. Ideally, power ~ clock speed, power ~ voltage squared, (max achievable) clock speed ~ voltage.
. At about 3 GHz it shouldn't be more than 60 W though.2. AFAIK, it's rather complicated but as a rule of thumb it's a square relationship. So over/undervolting has a much greater impact than over/underclocking.
Edit: found some formulas. Ideally, power ~ clock speed, power ~ voltage squared, (max achievable) clock speed ~ voltage.
-
tehcrazybob
- Friend of SPCR
- Posts: 356
- Joined: Wed Jan 16, 2008 8:56 pm
- Location: Council Bluffs, Iowa
- Contact:
It's actually fairly straightforward, but there are a few pitfalls to watch out for. First, here's your equation:
Now, that looks very simple, but there's a catch - the thermal design power for a processor is generally way higher than the actual power consumption. In order for that equation to work, the "stock" values should be the TDP, voltage, and frequency of the fastest chip available in that series, not of your processor. You can put in your particular chip's values as the "actual" values, and figure out how much power you're using out of the box, or put in your over/under-clocked/volted values to get your current consumption.
I'm not entirely sure how this works with Intel 45nm processors, since they didn't change the TDP when they changed the process; I'm not entirely sure any of the 45nm chips use 65 watts - but I'd still use the fastest one in the current 45nm range to start my calculations. The numbers are also slightly off when you take into account different cache sizes, but I don't think that difference is very significant.
As an example, in case things aren't very clear, we'll use my E6550, running at stock speed and 1 volt:
The fastest chip in the range is an E6850 at 3GHz and 1.25 volts. First, let's find the stock power consumption of my E6550 (still at 1.25 volts, but now only 2.33 GHz). Actual Power = 65 watts stock power * (2330MHz/3000MHz) * (1.25v/1.25v)^2 = 50.5 watts.
Now let's look at my undervolted settings (still starting with the fastest processor for simplicity, although now that we know the actual value for an E6550 we could just as easily use that number): Actual Power = 65 watts stock * (2330MHz/3000MHz) * (1v/1.25v)^2 = 32.3 watts.
For one last example, we can go ahead and calculate your power. Again, we'll start with the E6850: 65*(2.8/3.0)*(1.288/1.25)^2 = 64.4 watts. In your case, the higher voltage exactly offsets the lower clockspeed. With that overclock, you might not have much overhead, but you might want to toy with reducing your voltage slightly.
Now, that looks very simple, but there's a catch - the thermal design power for a processor is generally way higher than the actual power consumption. In order for that equation to work, the "stock" values should be the TDP, voltage, and frequency of the fastest chip available in that series, not of your processor. You can put in your particular chip's values as the "actual" values, and figure out how much power you're using out of the box, or put in your over/under-clocked/volted values to get your current consumption.
I'm not entirely sure how this works with Intel 45nm processors, since they didn't change the TDP when they changed the process; I'm not entirely sure any of the 45nm chips use 65 watts - but I'd still use the fastest one in the current 45nm range to start my calculations. The numbers are also slightly off when you take into account different cache sizes, but I don't think that difference is very significant.
As an example, in case things aren't very clear, we'll use my E6550, running at stock speed and 1 volt:
The fastest chip in the range is an E6850 at 3GHz and 1.25 volts. First, let's find the stock power consumption of my E6550 (still at 1.25 volts, but now only 2.33 GHz). Actual Power = 65 watts stock power * (2330MHz/3000MHz) * (1.25v/1.25v)^2 = 50.5 watts.
Now let's look at my undervolted settings (still starting with the fastest processor for simplicity, although now that we know the actual value for an E6550 we could just as easily use that number): Actual Power = 65 watts stock * (2330MHz/3000MHz) * (1v/1.25v)^2 = 32.3 watts.
For one last example, we can go ahead and calculate your power. Again, we'll start with the E6850: 65*(2.8/3.0)*(1.288/1.25)^2 = 64.4 watts. In your case, the higher voltage exactly offsets the lower clockspeed. With that overclock, you might not have much overhead, but you might want to toy with reducing your voltage slightly.