ACTUAL Intel Processor Power Consumption!

CharlieChan · Post by **CharlieChan** » Wed Jun 30, 2004 4:58 pm

MikeC wrote: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.

A quick browse of the design guide for the 875P chipset reveal that both the CPU and chipset requires Vcc. That means the 10W-11W I measured on 3.3V line of the 20 pin ATX connector is probably used for the memory and other low voltage components. That leaves ~10W, if I assume 50:50 split between chipsets and CPU that would mean the P4 2.0A draws 54W and the P4 2.4B draws 60W. At least this is closer to intels TDP figures.
It could also be true the dual 12V line is used to power other components on the motherboard such as the VGA.
Anyway, time to draw this whole excerise to a close - I need to get my computers back to folding otherwise hayesdb will start to PM me due to the drop in points

.

Petr · Post by **Petr** » Tue Aug 10, 2004 8:32 am

Hi!
I will explain how I calculated those power numbers. But first a little explanation of how CPUs work.

Modern processors use RISC core. x86 instructions are decoded to OPs and those are send to execution units in a way maximizing overall usage of those units. Athlon CPUs have 3 instruction decoders that can decode up to 3 instructions a cycle into OPs (3 or more OPs). Athlon can execute up to 9 OPs a cycle (3x ALU, 3x AGU, 3x FPU). Power consumption depends on how many OPs CPU executes in a cycle and on other factors like branch prediction power usage, FSB unit usage ect. AMD and Intel know something called "ideal instruction mix" what is a sequence of instructions resulting in maximum current draw (IccMax) and therefore max. power usage. This instruction mix is very unlikely to happen in common applications like Word, Xvid coding... but some very optimized code can come close to it (for example BurnK7 executes in six cycles 17 instructions, nearly peak of 18 instructions).

AMD and Intel set IccMax value according to ideal instruction mix what is also the design point for motherboard's power circuits. The max. power is IccMax * Vcc (one of basic laws of electricity). AMD's TDP values are Icc * Vcc, so they can be considered maximus. Intel's TDP is, on the other hand, based on typical Icc, not IccMax. Typical is Icc used while executing common applications.

If you have a 89W Athlon 64 CPU, this means you can expect something like 70W of power while doing Xvid coding. On the other hand when you have Intel P4 with TDP 89W, that means you can expect 89W of power while doing Xvid coding. Maximum power of Intel CPUs can be easily calculated by taking official IccMax value and multiplying it by Vcc. According to Voltage Regulation Down guidelines and P4 datasheet, P4 should be supplied reduced voltage when Icc increase so you can't just multiply IccMax with VID. Why I prefer comparing maximum powers and designing cooling solutions according to maximum powers is Intel does not guarantee power consumption won't be higher than TDP. My computed max. power is exactly equal of Intel's own numbers (see Xeon Prestonia datasheet that lists both TDP and max.).

There are other things you should be aware of:
1. Values from datasheets are for the worst case chips. TDP and of course calculated max. power is directly applicable to chips with highest possible VID and for highest IccMax. This may, hovewer, be different each piece... very much. For example, I had 1700+ Thoroughbred that is guaranted to about 55W. It was cool when overclocked to 2000 MHz at stock voltage. Now I have Mobile Athlon XP-M 2600+ officially rated at 47W and it runs hotter at stock in idle, that is at 2000 MHz / 1.45V ! This is because the XP-M has more leaky transistors, resulting in more static power (idle power). This difference can be observed between the same CPUs as well - the same as why one CPU overclocks better than other.

2. CPUs consume power even when idle. This power is not negligible. The newer the CPU, the more Amps it takes in idle. Prescott can, according to Intel's datasheet, take up to 40, 50 or 56A ! (depending on FMB / PRB class). If you want to calculate actual power usage, you must add some expected idle power.

3. Power supplied to CPUs can come either from 12V Power or from Main Power of from combination of both. Newer boards usually use 12V Power (LGA Prescotts should use only this one as new ATX12V standard version 2 was specified expecting 14 or 15A only for those processors), sometimes with combination with Main Power.

4. When you load CPU, then also chipset and RAMs can be loaded more (depends on application). Calculating thermal power of CPU only is very difficult.

Post by **MikeC** » Tue Aug 10, 2004 8:53 am

Hello Petr,

You must be the Petr Koc (alias Eagle) who created the CPUHeat & CPUMSR projects homepage that began this whole discussion! Wecome to SPCR. Perhaps you noticed the flow of traffic from SPCR to your site that led you to investigate and find this... I hope you return often to contribute to the discussions here. Any new interesting projects on the go?

-- back on topic --

Much thanks for your explanation of how Max power of Intel CPUs was calculated. It makes perfect sense. But having read through your whole post, I wonder if by...

Calculating thermal power of CPU only is very difficult. (your last line)

... you mean measuring thermal power of CPU only is very difficult.

Petr · Post by **Petr** » Tue Aug 10, 2004 9:22 am

Thanks!

Yes, I mean "measuring". Sorry for any inconvenience.

Projects - I am now working on a PowerNow! utility but it takes much longer than I expected. We are two people who work on this and just do not have enought free time in right time. I hope I finish it in few weeks.

CharlieChan · Post by **CharlieChan** » Sat Aug 14, 2004 7:21 pm

Hi Petr,

Thankyou for the explaination. I understand how you derived at the results but still think that an attempt should be made at measuring the power. It is not that I doubt your calculations but the figures supply by both AMD and INTEL. As an example AMD specifies the XP2400+ has a maximum Tpower of 68.30W (in fact 5 CPUs from 2400 - 3000 has the same Tpower!) but when I measure the power at the wall when using this CPU in a diskless folding farm I get 120W. Nothing usual, until I measured a P4 2.4B at 100W doing the task. The XP used 512 of DDR and the P4 512 of RDRAM, although the MB were different both were measured using the same PSU (SS300FT). In my mind these results raises many questions especially when your calculations state the P4 has a maximum power of 69W. Unfortunately I do not have the time to device a method to estimate the power consumption of a CPU.

Charlie.

Petr · Post by **Petr** » Sun Aug 15, 2004 1:53 am

Maybe Intel has tighter tolerances? Considering both CPUs are performance equal (2400+ vs. 2.4 GHz) maybe just in this application P4 can profit from it's architecture to do the same task with lower power. Have you tried other applications, ideally heavily optimized for both platforms? I am thinking of Prime95. I think the case is similar as with Prescott. I saw first Prescotts running 70 degrees C and 50 in idle. But Intel's papers say Prescott is 103W max. power while Northwood is 96W (... and running 60 degrees). Such a small power difference can't make for such a huge temperature difference.

How to measure power consumption - I have an idea. Because nearly all power supplied to CPU is transformed into heat, it may be possible to put CPU's heatsink into some box, run Prime95 for 10 minutes and then calculate temperature difference of air inside the box (before and after the test). Air must have some thermal capacity so if you know dimensions of the box and know the thermal capacity, it is possible to calculate how much energy was transfered into air. And if you know time then calculating power is easy.

deebass · Post by **deebass** » Fri Sep 03, 2004 6:32 pm

Hi!
So long since my last post over here. I'm very busy at these moments, but sometimes I keep on coming over here to have a quick look on what's boiling

Nice and interesting thread.

Very precise post Petr. I've always taken into account that when considering thermal issues. I have to say that I still think that VRD guidelines (and its variable maximum current capabilities in different revisions of the norm) is the main reason of some "fatigue" under certain circunstances in P4 systems. That usually makes (and Intel promises) a P4 CPU not to pass its TDP figure for long time (it's OK for a minute, but not for an hour).

Besides, about P4 power discussion, Intel used to provide maximum power figures until Willamette. Extrapolating those power figures have usually worked quite well for me with Northwood (as there have been no main changes in architecture with this core). I haven't tested any PresHOT (and I won't, I think

), but I guess it could be still a good first aproximation (although getting that "ideal mix of instructions" may be impossible

).

For those interested, I recommend you study carefully Erol's "Processor Electrical Specifications" Web page (http://users.erols.com/chare/elec.htm). The key clue is on P4/P4 EE notes. I hope I've been useful and this haven't been discused before.

Sorry for my horrible English. It's been so long since I don't make myself understand in English.

Greetings from Spain.

jojo4u · Post by **jojo4u** » Tue Apr 05, 2005 2:36 am

Hardwareluxx.de shows the power usage of a full system equipped with the P EE 840 and the P4 D 840. Under load it's 7% more for latter and ~17% more for the P EE 840 compared to a P4 570J. Idle sees around 5% more.

halcyon · Post by **halcyon** » Wed Jan 04, 2006 2:15 am

Intel 955Extreme Edition (Presler core, 65nm, 3.46Ghz):

- 61.6W at idle (Windows XP)
- 144.8W at load (Prime95 load - both cores)
http://www.lostcircuits.com/cpu/intel_p4-955/11.shtml

- 156.2W at full load (S&M burn - both cores)
http://www.xbitlabs.com/articles/cpu/di ... ler_8.html

Natronomonas · Post by **Natronomonas** » Mon Mar 12, 2007 5:02 pm

Anyone know of some actual C2D power consumption numbers?

I found some here:

http://www.lostcircuits.com/cpu/intel_core2/7.shtml
http://www.lostcircuits.com/cpu/intel_core2quad/5.shtml

but I was looking for e4300.

(not TDP, actual consumption...)
Thanks...

accord1999 · Post by **accord1999** » Sat Mar 24, 2007 1:37 pm

Natronomonas wrote: but I was looking for e4300.

(not TDP, actual consumption...)
Thanks...

http://www.behardware.com/articles/652- ... e4300.html

Natronomonas · Post by **Natronomonas** » Mon Mar 26, 2007 5:50 pm

That's great, thanks!

maria.johnson9 · Post by **maria.johnson9** » Mon Oct 27, 2008 7:01 am

Noticeably, DFI took a good care of the efficient CPU power consumption management

gb115b · Post by **gb115b** » Fri Sep 11, 2009 6:01 am

any updated version of these charts?