Hi, all silence lovers. Did some energy consumption measurements of my home setup. These were the objects of interest:

1) Biostar G31-M7TE (1.18V on GMCH)
2) Intel E5200 2,5GHz (1.1250V)
3) Patriot EP 2x2Gb DDR2-800 CL5 (1.85V)
4) Force3D HD4670 750(1.3V)/2000MHz(1.8V)

Pls note, that both ODD and HDD drives I have supplied bypassing our field of test. So we will only receive results of motherboard, CPU, RAM, and video. Measurements were performed in DC area, so we must add to our findings some losses based on imperfect efficiency factor of the PSU too.

Measurements were done by the means of an analogue amperemeter (well, galvanometer plus shunt resistor, actually) I used to use in my homebrew experiments. I believe it has enough accuracy for this one too. Of course, I have a digital multimeter, but here I took the analogue device, cause it's good for showing an average value.

Now conditions. The idle is idle. No special peculiarities here. Intel's SpeedStep enabled -- multi drops from 12.5 to 6, ATI PowerPlay functioning -- 750/2000MHz -> 165/250MHz. There is also nothing launched in Windows besides the needed applications such are Linpack, Prime 95, and ...Flatout Ultimate Carnage. The choice is conditional on the state that we must generate as much heat as possible. E.g. Prime 95 just loves to load the +5V rail, and Linpack is unsurpassed for +12V, doing this better than well-deserved Prime 95. And concerning video part nothing can load so much your compie as, yes, a game. So in 3D load scenario I was launching the Flatout game (~35-40FPS on maximum detail). Well, so much for conditions, let's go to figures (are given in Watts):



And some additional information.
1) Take a note that the present G31 GMCH is a bit undervolted (done through simple pencil trick) -- 1.18V instead of 1.28V stock by Biostar (and 1.25V by Intel). This drop down of 0.1V brings 1.3W economy in idle, and 2W in load (+5V rail impact only). I just forgot to remove the mod before our test.
2) If we pull out one RAM stick, we earn 2W in idle, and 4W in load (+5V rail impact only).
3) Raising the default frequency of E5200 up to 3GHz (on the stock VID 1.1250V) gives consume increase by 0.2W in idle, and 6.4W in load (+12V rail impact only).
4) If we decrease the E5200/2.5GHz VID to a bit low value -- 0.9250V -- it gives us a drop of 1.8W in idle, and 11.4W in load (+12V rail impact only).
5) The +5Vsb rail consumes 1.25W in standby, and 0W power on (insignificant -12V I have skipped).

That's a very interesting study, especially for people who want to develop extremely low-noise (voltage-wise) computer. Although you may not be getting the right target audience here as people are mostly interested in noise and overall consumption.

Based on my limited understanding, both the CPU and Graphics card, as well as many chipsets in the motherboard depends on the 12V only, as VRM is used to generate the various low voltages from 12V and these switching VRM are more efficient when driven from a 12V.

Only a small number of low-power devices, e.g. sound chips in motherboard depends on the 5V, as they use linear LDO regulator to generate low noise 3.3V to supply to these sensitive chips, and linear regulator is more efficient when driven from a 5V to 3.3V.

I am a bit surprised that the RAM depends on the 5V as I thought they are also driven by VRM which depends on 12V.

As the CPU should only gets its power from VRM which draws on the 12V, I wonder how the prime could stress on the 5V. Are you using the stress memory option in Prime 95?

I have an Asus G33 P5k-vm board, I am also interested in lowering the GMCH, any idea if that could be easily done?

Nice choice of parts, less power means less heat so easier to cool. Less than 100W load and 50W idle is awesome!

I would be tempted to keep the e5200 at stock, but see how low you can drop vcore. You may even find with a low enough vcore you could run the e5200 with a passive cooler!

Is ATI powerplay a Windows only feature? I run ubuntu also you see, and I'm also tempted by the 4670, but only if Powerplay actually works...

Finally, is there much difference in power draw between motherboards? I'm running p965 right now, and am reluctant to change...

Yeah, +5V was used both for chipset and dimms supplying. I think this was done to make the PSU's load more balanced as long as we have 5+ and 3.3V rails. Why not use them, if they do exist yet. The only 12V+ supply for motherboard and devices looks excellent, like a dream, but the PC world is very inertial and conservative.
I was using the mode called "Blend (tests some of everything, lots of RAM tested)". Don't forget that chipset uses +5V rail for feeding its internal circuits, as the DRAM too.
If there will be a small lurking resistor that can be painted over with pencil's graphite... I did some search for hires pics of the G33 P5K-VM, and I have some doubts about the easy way to do undervolt on it:
http://www.unitycorp.co.jp/asus/motherb ... photo.html
http://www.nix.ru/include/view-photo.ht ... 4&pid=2245
Can't see inductances near the GMCH, and they are sign of PWM control (the easy way to do the mod). I think, that this board uses linear converter, so we must search for smth. like LM324... Oh, that will be a bit intricate, I feel

Some problems here. I can't find the documents for STM PWM chip used for CPU Vcore management. I want to have the posibility to do smooth adjust, not step-by-step by VIDs as it is rather tricky (pulling out cooling system, managing with scotch, wire jumpers, that's mess).
Search the net for ATI driver for Linux capabilities. It depends only on driver, I believe.
There is a sticky theme in the forum "TDP list for Intel chipsets", look there

do you need to do something to enable powerplay, or does it operate all the time automatically (I have a HIS 4670)?

Thanks.

Your HIS has full PowerPlay support, so don't worry

Generally it depends on the manufacturer of HD4670 you bought (it's BIOS), but usually you don't have to do anything to enable it, as it already running if you get in Windows (or Linux?) with the driver installed. If the PowerPlay is still missing it can be added via BIOS editing. Besides the RBE I'd recommend some other great stuff: Rivatuner, to watch the clocks, and, if you use WinXP, this ripped off ATI Radeon driver (only 7.45Mb for v8.11 with control panel, and no annoying NET Framework!)

Here is some homebrew development info for Biostar G31-M7 TE owners on the hardware voltage adjust.

G31 chip:


DIMMs:


FSB Vtt and ICH7 chip (ICH7 has a separate 1.05V and 1.5V core supply):


...

Now let's see where we can measure what we are doing there to the Vich, as there is no info in BIOS about this thing. Btw, all BIOS monitor readings are (surprisingly) very accurate.

Check point for Vich:


Vdroop mod can be done also, but it concern a physical damage(!) of the board (you must disjoint two pins of CPU VRM chip soldered together).

Kinda FAQ:
Q) How can I set the voltage I want?
A) You have to paint over the corresponding resistor with a thin regular pencil. Do about five passes then check in the BIOS monitor (or see check point pic) if the desired level has been achieved. Use this information with caution. Do not forget that you may apply too much voltage and this will take out your hardware to heavens!

Ksanderash: Nice results. But...

Can you post consumption with all components (the whole system)?
Cpu+mb+Hdd+Psu+DDR+VGa etc? And write type ...

I'm interested about consumption Gigabyte GA-EG31MF-S2 and BIOSTAR MB G31-M7 TE. Can you tell me the different?
And can you post your 3dmark 2005 score with HD 4670 ?

Thx a lot.

No problem here, but a bit later, OK? I must calibrate my (not TrueRMS) multimeter to show overall AC wattage as I don't have this handy thing, 'killowatt" or something

No, as I don't have one. Have you seen this review, Foxconn vs Gigabyte:
http://www.tomshardware.com/reviews/int ... 39-13.html

No, sorry, I can't, I dont' use it. But my 4670 is absolutely equal to others 4670 in this mean. You can see the results over the Internet.

Ksanderash: OK, i will be waiting for.

The review i saw. Nice Foxconn at idle, but hardly get it in my location.

The Hd4670 - i know is equal, but me interesting the real benchmark with G31 chipset because i need to know if helped me change mb+cpu with better performance. Doesn't matter , soon i will get packet with that Gigabyte G31 mobo with E5200 and i will test it.

Thank you for posting.


Edit:
Today arrived cpu E5200+ Gb EG31MF-S2. Mb looks nice, small micro Atx form, but have everything i need. Cpu looks good to, low noise box cooler. I believe low consumption cpu, heatsink is small like for 1Ghz cpu .
Now i'll be waiting for consumption check meter. When arrived i will post results. I hope will be the 40 w at idle Wink
I wanna check first consumption my current sys with that meter.

Overclock i wanna just try, and i guess 3 Ghz is enough for me.

ex.treme
That's great. Will wait for your results

E5200 has remarkable energy efficiency. I have disconnected the cooler at all, and the temp at 2.5GHz/0.9V don't go higher than 50C in usual Internet browsing! To say honestly there is a blow away 120mm PSU fan at the top of it, and it does the most of the convection.

...

Some problems with my board -- it appeared that it has no physical MAC :-/ So I RMAed it, and now waiting for another one.

UPD: Ha-ha, RMAed in vain, cause me being silly as a goose couldn't suspect that I'm installing the wrong driver

The OEM heatsink that came with my E5200 is so dinky. Haven't had time to test it much on my TUL(Foxconn) G31, but I did undervolt and overclock it in the BIOS to 3.2GHz with no issues. It is runnning fanless with a Coolink BAT heatsink. As I thought, the TUL board doesn't support 1:1 divider with this CPU, but I will try to get the BSEL mod working.

I mounted the system.

E5200+ GB EG31MF-S2, 2x1Gb Kingston DDR2-800 dual, WD 320Gb BEKT 2.5', GB HD4670, 350w Fortron GLN.

Consumption checked with Energy check 3000.

V core - 1.000v at bios , ( with def. v core consumption) Dynamic Energy saver (DES) helps drop some wattage, but undervolting its better.
Max w
1.00v ( def voltage) DES
Idle - >1.2 Ghz - 54 w , ( 57w ), - DES - 53w
Load super pi - 2.5Ghz - 65 w , ( 80w) , DES - 74w
Load Orthos stress - 2x cpu - 72 w, (88w) ,
Load 3dmark 2001 nature loop -2.5Ghz- 108 w, (120w), DES 118w

I will try set vcore 0.9v like you. But i guess dont help me get 40w at idle
Maybe with igp

EDIT: with Igp - idle 45w. Orthos stress cpu - 63w

At bios with 0.9v windows freeze

ex.treme

Pretty nice! Do you have an optical drive there? I can't wait to see my board from service to compare with yours I have a Fujitsu 160Gb 2.5" drive and can hook it on to, instead of 3.5" WD3200AAJS, to equalize conditions.


And what CoreTemp tells about the CPU's factory VID? Mine has rather low, 1.1250V. I have by the hand one with 1.1875V.

There is option C3E or C4E that drops 2 watts idle when enabled.

Using 1 x 2gb a-data module, instead of 2 x 1gb, undervolting that 800mhz memory to 1.65 or 1.55 instead of 1.9 does drops even more and it is stable! that is at least 10 watts easy including mch undervolting 0.95v... is real with eg41mf(full 16pcie) probably too.

Mine is ep45-ds3/e5200 @ 1.2ghz/0.8v @ 4watts full occt load cooled by ZM-VHS1 .

Ksanderash:
i have optical drive DVD-Rw installed and mounted.

Core temp shows : low idle - 1.125v- 1200mhz, load- 2500mhz 1.2675v lol ?

I don't think is correct. I have set up at bios 1.ooov vcore.

I remember an article, where the graphic cards' draw was measured using "a modified motherboard". Seems rather difficult to modify...
Ksanderash, you inserted resistors in the wires from the PSU to the MB, correct? What was the max voltage drop, some 50 mV? Quite a low Ω resistors. Did you change the resistor when the current changed? Any chance of switching the power on without one voltage, would that result in damage?
They are more efficient from 5V, but 300V to 5 V is less efficient than 300V to 12V.

Yeah, you must cut positive PCI-E power lines from motherboard to slot-connector. It's possible, but very tricky. Therewith there are several pins, not one or two.
I was using an integrated in the multimeter shunt (resistor). There is a quite low impendance -- about 1mm thickness copper wire. Thus no serious voltage drop having so much impact on the measure accuracy.
So, I measured not volts, but amperes. And to bring down the number of applicable power lines (there are 24 contacts in ATX2.2, as we know), I combined equal voltage outputs into one. And as a result only three rails: +3.3V, +5V, +12V. Now it's much easier to deal with.

To the board? I don't think it will cause a permanent physical damage (we have only RS232 buffers on bipolar supply), however I would't try to prove it

Ksanderash, can you please post details on how to do vdroop mod for Biostar G31-M7 TE?

That is a bit difficult task to proceed. Locate a chip named L6703 (L6714 nearer analogue). You must disconnect the pin 21(Droop) from the pin 22(FB), and connect it to pin 19(SGND).

Excellent data.
Are you sure about 5V load it seems lower then mine I also have a G31 but load on 5V is higher an I know you undervolted I did the same using bios settings as -0.15V it seems that Biostar is better than Gigabyte on low power motherboards. you can see my measurements if you go to the link that I have in signature.
The G31 is quite old now and I will use the Gigabyte until a better low power MB with integrated video will be available or maybe I will go directly to core i3.

Thanks! Absolutely sure. I'm very pedantic when it comes to figures ) And, yes, this G31 board has a way modest current consumption. Only ATI/AMD chipset powered boards can go lower but I stick with Intel CPUs since Core architecture appeared.


[UPDATE]
Damn, I've broke with my lovely G31 board ) Got to buy another...