A new link added to the Information category of SPCR's Web Links, called CPUHeat & CPUMSR ( http://www.cpuheat.wz.cz/html/IntelPowerConsumption.htm ) provides tables of ACTUAL Intel CPU wattage under max stress such as CPUBurn:

The tables are eye-openers. examples:

P4-2.8 Northwood: 79.3W
P4-2.8 Prescott: 100.4W
P4-3.2 Prescott: 115.4W

PS -- Thanks for Steve of ARM Systemsfor this invaluable tip.

Ooh, nice find. Thanks! I bookmarked this and downloaded the tables.

After a quick look at his tables, it looks like his max wattage numbers are typically around 10% higher than Intel's "TDP" numbers.

How much does the Top-O-Line AMD's draw? :S

At computerbase.de - the guys who already provided VGA power draw figures - is a comparism of the AC power draw between Prescott, Northwood, Barton, Athlon 64. Prescotts is the all-time looser here...
Take the idle-figures of the Barton with a grain of salt since disconnect seems to be disabled.
Page 3 shows the energy costs in Euro with the price of 20euro-cent per kw/h.
Page 4 first table is a custom performance rating based on prior cpu tests. Next table divides performance rating by maximum power draw.

400-Watt-PSU
Asus GeForce FX 5900 Ultra
IBM 40 GB Desktar 120GXP
TwinX1024(RE)-3200LL
Pentium-4 Asus P4C800-E Deluxe
Hyper-Threading activated
AMD Athlon 64 MSI K8T Neo-FIS2R
Athlon 64 FX Asus SK8V
Athlon XP Asus nForce 2 400 Ultra

http://www.computerbase.de/artikel/hard ... ozessoren/

jojo4u --

Those are great tables! Thanks for the link. I think I'll have to find a way to post this info permanently here...

My only complaint about the tables/graphs is that the order should have been reversed, with the lowest power dissipation CPUs on top. Then even people who don't read and only scan would know the Prescott is the loser.

Computerbase.de does it again. This time, they test how power draw of Athlon 64 3800+ scales while overclocking. Same components as in post above.
http://www.computerbase.de/artikel/hard ... mrechnung/

The new socket 775 is out, and computerbase.de updated the figures.

http://www.computerbase.de/artikel/hard ... attform/6/

All CPU here, except the P4EE are Prescotts.
Although the new 3,2GHz Precott in C0 stepping (540) does fit the "Platform Compatibility Guide 04A (PCG 04A)" with a maximum TDP of 84W, there are no significant saving compared to the 3,4/3,6GHz one.

Buried in Tom's Hardaware's massive Intel 775 mptherboards article is a page w/o notes called Processor Power Consumption.
http://www.tomshardware.com/motherboard ... 75-49.html

This page displays two charts: CPU power consumption at idle and while running Prime95. It seems to be CPU power draw only. No information whatsoever on how this data was obtained.

All the numbers look low compared to others discussed in SPCR -- and in this thread. 62.6W for a P4-3.4E just seems plain wrong. So does 52.9W for an Athlon 64-3400+.

But given THG's unresponsiveness to queries about their articles, we'll never know how this info was obtained, who from, etc. I'd take it all with a huge sprinkling of salt, but it's still interesting.

The discussion thread can be found here:
http://forums.silentpcreview.com/viewto ... highlight=

my extra complains:
no disconnect, no cool'n'quiet(athlon 64 needs more at idle then northwood)

I can't speak to the AMD results, but it's fairly common knowledge that Prime95 doesn't fully load the P4 CPU, even if the idiots at THR actually ran two instances of it. Doesn't that negate the value of that entire test?

Some new shocking figures by the german magazine C't:
figures idle/3DMark in Watt.
P4 560 3,6Ghz, i925x, Nvidia NV45 protoype, unknown PSU: 156/316W
P4 3,4EE, i875P, Nvidia Geforce 6800 ultra, unknown PSU: 151/295W
A64 3800+ 2,4Ghz, K8T800 Pro, Nvidia Geforce 6800 ultra, unknown PSU: 121/232 with manual Cool'n'Quiet: idle@93W

The contrast against the 3800+ is shocking.

Too bad they don't specify the PSU, jojo4u. If they did we could get a handle on how much DC power is reallly being demanded.

I find it difficult to accept their calculated results without empirical data. They could have measure the current drawn by the CPU using a DC clamp meter. I setup a test environment to measure the current drawn by a P4 2.0A and a P4 2.4B. The power dissipated by both CPUs were 20% less than the calculated results.






The setup comprise of CPU, MB, RAM and Seasonic SS300FS running a diskless version of Linux. The readings were taken after running burnP6 for about 5 mins.

CharlieChan --

Your results then suggest that actual power draw of P4s is lower than Intel's TDP -- which we know is simply not true; Intel says plainly that actual power can be higher.

Besides, how were you able to isolate the power to the CPU section alone? What was your load? And how did you measure AC power? (Not nitpicking -- curious about your test setup).

AC power was measured using a Power Meter similar to the Kill-A-Watt meter you use. DC current was measured using a CLAMP METER - this meter uses the HALL effect to measure current in a wire without breaking it. The current to the P4 is supply by the 4 Pin connector on the motherboard.

You know, reading your posts reminded me that in fact, I have a Fluke 36 Clamp Meter as well! (On loan from PowerGyoza, who used to be very active at SPCR in the past...)

Dug it up, caibrated the Ammeter to zero outside the house, then measured the 12V lines on the test board running a P4-2.8 (533). With CPUBurn, it registered 6.6A. Measured 11.9V across the 12V terminals (where they connect to the board).

6.6 x 11.9 = 78.5W

Pretty darn close to the 79.3W predicted by CPUHeat & CPUMSR.

Perhaps not all of it.

That is interesting, my measurements were taken using Linux but just to be sure I when and took another set of measurements from a WXP machine running two instances of burnP6 on a 3.0C. The current recorded using the min/max function was 6.29A. I could not take the voltage reading due to the surrounding components but motherboard monitor recorded 11.49V.

6.46 X 11.49 = 74.22W

I do not know where the motherboard and other components draws their power from which is why I initially try to measure using the minimum number of components to get a OS working.

Tried it with a P4-2.53 (533) as well: 3.8A x 11.98V = 45.6W

Way lower than predicted by either TDP (62W) or the other (75W)!

Looks like this method has its problems...

1) have not checked batteries in clamp meter, which as been unused for a year. Need to do that and retest.
2) we may be seeing the power efficiency variances in CPU samples that the AMD engineer went on about.

That could be the case but the measurement taken when the CPU was idle was



These results would indicate most if not all of the current drawn by the CPU is from the 4 Pin connector.

The reason I took AC measurements and ran a minimal system was ensure the measurments taken with the clamp meter were reasonable.



Very likely, I record from my undergraduate course in digital electronics (~10 years ago) that most complex chip designs have redundancy build into them. This would be prudent otherwise one faulty transistor or module could render a chip useless.

Changed the battery to a new one, measured the 2.8 again. This time I only got 4.2A!? 4.2 x 12 = 50W. Way lower than either TDP or the other.

System power draw was 131W (AC) - efficiency of the Seasonic SS300FS PSU powering this system is 72% at this power level (from previous testing results). This means it was delivering 94W to the system:

Intel 845 board
Barracuda IV 40G drive
GF MX400-64 VGA
512MB DDR3200 ram
Panaflo 80L (0.1A)

I suspect those will not draw 43W, VGA & HDD at idle. Not sure of RAM draw power regardless...

Does not seem right.

I guess I could measure the DC power to the board & the HDD.

The motherboard and RAM should draw at least 20W, seagate rates the barracuda at 9.8W idle - will the VGA draw 13W? If we assume the VGA only draws 5W then the CPU draws 57W.

Oops! More user error.

I was measuring only the +voltage leads the first time (which is what you are supposed to do) and then promptly forgot to repeat that for the rest. So remeasured everything --

6.6A on the P4-2.8 = 78.5W
5.2A on the P4-2.53 = 62W

The latter is dead on the TDP prediction. I'll have to round up all the P4s I have and try it on one platform.

If both statements are true it means the rest of the system (MB, VGA, HDD, FAN) only draws 15.5W .

With everything at idle, isn't that reasonable?

From my playing with wattages, 15-20 watts for mobo/ram/hdd/vga is about right, at least with the low-end vga that we're talking about.

The short answer is I don't know. I am trying to find out by running these little 'experiments' and I am making certain assumptions from the results. One of the assumption I made was that the CPU draws all the current from the 4 pin connector - this could be wrong. I took measurements when the CPU was idle and when the CPU was loaded, 99% of the difference between the readings occur on the 4 pin connector. I was running a OS over the network with no hard drive, floppy, cdrom and was using onboard VGA. The OS ran in console mode like MSDOS so onboard VGA was not used. The power drawn at the 20 pin ATX connector during idle was 20.93W and the power drawn under load was 20.58W. I also took measurements without a CPU and the power drawn was 16.27W.

It is possible that the OS is preventing burnP6 from running at max priority hence the lower than expected results.

This is were I got the idea of using a clamp meter - I had originally thought about building a circuit with 10 meters to measure the current of a motherboard. Good old GOOGLE can to the rescue.

After a long chat with Ralf Hutter, we have come to the conclusion that the dual-12V lines are not the only source of voltage for the P4. RH mentioned that with earlier lower powered P4s, they actually could run without the dual-12V line plugged into the motherboard; just that when you try to put a heavy load on the P4, it tends to get flaky. They were intended to be auxiliary 12V lines -- ie, a way for the board to get more juice directly when needed.

Hence the widly varying results...

I am sure there is a tech doc somewhere on Intel's site that would give precise info on this, but there's no way I'm going to dig through them right now.