Core2 idle consumption

tehcrazybob · Post by **tehcrazybob** » Fri Apr 18, 2008 5:14 pm

I've got a Core2 Duo E6550 in a Gigabyte GA-P35-DS3L, undervolted to 1.1v in the BIOS, which results in an actual core voltage of 1.056 as read by CPUz. I've also got all the power management options in the BIOS enabled. I've done some testing with a kill-a-watt and found that after correcting for the power supply efficiency, the difference between idle and full load is 20 watts.

By consulting several overclock calculators and the information provided by SiSoft Sandra, I've determined that my processor's full-load power draw is somewhere between 30 and 35 watts. This means that my chip is consuming between 10 and 15 watts at idle, but this seems rather high. X-Bit Labs tested this processor and concluded that it draws only 5 watts at idle.

Why is theirs idling so much cooler than mine, and what can I do to close the gap? Or are my estimates incorrect, and my processor is only drawing 25 watts at full load?

yuu · Post by **yuu** » Sat Apr 19, 2008 3:28 am

xbitlabs "the actual CPU power consumption disregarding the losses in the voltage regulator circuitry"

but I don't see this board having big vrm losses
here:

E2140 333 * 7 in a Gigabyte GA-P31-DS3L, u/d to 1.1v in the BIOS,

idle 98W / load 125W with orthos, tat +1W

the lowest i do

1600Mhz, 0.89V 91/103
1866Mhz, 0.94V 96/114W

the efficiency of my power supply is ~ 75-80%,

if you have 70 and 90W at 60 and 75% efficiency

you are looking at 62-47, that is the same 25W difference i have

tehcrazybob · Post by **tehcrazybob** » Sat Apr 19, 2008 11:07 pm

I apologize, but I don't quite follow what you're trying to say. To make sure I've gotten myself across clearly, here's a bit more detail on the situation:

My computer runs at 93w idle and 117w full CPU load, measured at the wall. With a Seasonic S12-II 330, I should be right at 80% efficient at that kind of load. That means that my computer is drawing 74 watts idle and 94 watts under load in DC from the power supply. That means that between idle and full load, my processor changes by 20 watts.

I've seen that this processor is supposed to draw 5w at idle, but I've also determined that with my undervolt I'm drawing between 30 and 35w at full load. This isn't a 20w difference between idle and load, though, which can mean one of two things. Either my power estimates are wrong, and my full-load consumption is a mere 25w, or my estimates are correct and I'm actually drawing 10-15w at idle. I'm wondering which of these is more likely, and what I should change if my idle power is higher than it should be.

If it helps, I've noticed that in CoreTemp, my processor state is always C0. Shouldn't this drop to C1 or C1E at some point? The multiplier and voltage drop as they should at idle, so is CoreTemp just not noticing this, or should the processor be doing something further? If so, how do I enable that (minimal power management is already selected)?

yuu · Post by **yuu** » Sat Apr 19, 2008 11:40 pm

all that matters is that xbit labs results exclude VRM losses that could be 10W at idle for all i know, that is 30% efficiency, what do you know they measured it and substracted it, they didn't share this valuable information with us.

i hope spcr would enlighten the matter one day on everything.

i speculated about your psu. but it's still very close to what i have.

xan_user · Post by **xan_user** » Sat Apr 19, 2008 11:43 pm

I think one issue your seeing is the fact that PSU efficiency is not constant at different loads. 80% at full load does not mean 80% at idle.

Why are you trying to isolate only the CPU wattage anyway?
Isn't the most important number the actual power consumption of the whole rig, not just the CPU?

SebRad · Post by **SebRad** » Sun Apr 20, 2008 2:14 am

Hi, funnily enough I noticed just about the same results the other day with my system following testing with a power meter I bought. (Special offer in B&Q for Â£10)
E6600 (2.4GHz), Asus P5B-E plus, 2x 1GB, X1950pro (600/750), WD 5000AAKS, DVD, DVDRW, NeoHE 430.
Under-volted to 1.1v in BIOS (like OP) I saw Idle (6x266 = 1600MHz) of 100w(AC) and load (Folding@Home SMP) of 120w (AC). PSU ~80% efficient as 240V AC here.
Orthos and CPU Burn will produce higher load figures.
I did also briefly test with an S12-II 330 and the W figure was very similar, maybe 1-2w less.
Interestingly firing up ATI Tool 3D cube thing suggest the video card can add 50w to the system, or twice the CPU load!
Overclocked to the max (3.6GHz, ~1.4v Vcore actual load) and under full CPU and GPU load I can get to ~250w AC under load.
I also discovered that, as I thought, my 21" CRTs use ~100w each and also that my amp use 30w! I'll be making sure to turn the amp off when not using.
Regards, Seb

tehcrazybob · Post by **tehcrazybob** » Sun Apr 20, 2008 10:32 am

xan_user wrote:I think one issue your seeing is the fact that PSU efficiency is not constant at different loads. 80% at full load does not mean 80% at idle.

Why are you trying to isolate only the CPU wattage anyway?
Isn't the most important number the actual power consumption of the whole rig, not just the CPU?

I checked the SPCR review of the 380-watt version of that power supply, and in that range it varies from 79.5% to 81.5%. I believe 80% is a safe figure to use for both states, and a single percent difference is only going to shift me a watt one way or the other.

I was only trying to isolate the CPU wattage because the math didn't work out the way I'd expected it to. I know that overall consumption is what matters, but if you're trying to lower the total it doesn't hurt to know which components are causing it. As SebRad noted, modern graphics cards outpace modern processors by a wide margin. Strange then, that the heatsink on the card is so much smaller than the one on the processor.

Anyway, thanks to everyone for the replies. I might poke around my BIOS options a bit more to see if I'm missing something, but otherwise I'll just blame the losses in the VRMs and leave it at that.

astrayan · Post by **astrayan** » Sun Apr 20, 2008 9:05 pm

Your main problem is that you are subtracting a large number from another (to derive a small number), and assuming you know the error uncertainty of both numbers. If you did Physics at school, you'd know that was a no-no. The total power estimates can be way out, and conservative; vary from chip to chip.

I can't reconcile TDP of this Athlon Winchester (67W) with what I see on my power meter. I run iPrime max CPU stress, and get:

64.5W idle (0.875V) -> ~1W (baseline)
====================
80W (1.15V) full load 2GHz-> 12.4W
100W (1.35V) full load 2Ghz-> 28.4W

(This is thru SS-430HB PSU - say 80% efficiency)

This chip's max is 28.4W, even when I use full spec voltage. It's supposed to be a 67W chip. Just to give the environment a boot up the arse, AMD run it .2V higher than it needs to be, wasting 16W.

This is why reviews comparing AMD and Intel really shit me. None of the twits ever deal with undervolting, and how much you can do it on the reviewed chips.

How do I know what the idle power of a CPU is? I take off the heatsink and feel it with my thumb for a bit, and guess how many watts are there. That's another thing that shits me about CPU reviews. They never put on a custom heat sink that measures the real heat wattage. It's always done through some stupid motherboard, with a huge GPU stuck on it, and a hopeless PSU.

My rule is : any CPU that heats up when idle with no heatsink, is not idling, but doing satan's work in secret: possibly cracking your bank accounts and sending out sly information to big brother

The amusing thing is that a Winchester can run "half speed" (1Ghz), and crunch video at 60% full speed, and uses about 3.5W. A 130nm Newcastle core I just got seems to do the same at 0.825V and uses 5.5W.

jessekopelman · Post by **jessekopelman** » Mon Apr 21, 2008 1:36 am

tehcrazybob wrote: I've seen that this processor is supposed to draw 5w at idle,

5W at C1E, not idle. Idle is not continuous C1. The only way you could be in continuous C1 us to halt the OS itself. There are applications you can run that will force your CPU into C1 state more often during idle. Do they work better than Windows inherent management . . ?

tehcrazybob wrote:If it helps, I've noticed that in CoreTemp, my processor state is always C0. Shouldn't this drop to C1 or C1E at some point?

It can't drop below C0 if CoreTemp is actively updating. In C1/C1E, the CPU is doing nothing at all.