All numbers are maximum TDP taken from Intel.

IOH:

X58: 24.1 W


MCH:
(Note: Idle Power is based on a typical part in system booted to Windows OS with no background applications running.)


X48: 26.5 W, 12.3 W idle, with 333 MHz FSB
P45: 22 W, 9 W idle, with 333 MHz FSB
G45: 24 W, 9 W idle, with 333 MHz FSB
P43: 22 W, 9 W idle, with 333 MHz FSB
G43: 24 W, 9 W idle, with 333 MHz FSB

X38: 26.5 W, 12.3 W idle, with 333 MHz FSB
P35: 16 W, 5.9 W idle, with 333 MHz FSB
G35: 28 W, 11 W idle, with 333 MHz FSB
Q35: 15 W, 6.5 W idle, with 333 MHz FSB
G33: 14.5 W, 5.75 W, with 333 MHz FSB
P31: 15.5 W, 7.6 W idle, with 266 MHz FSB
G31: 15.5 W, 7.4 W idle, with 266 MHz FSB

G965: 28 W, 13 W idle
Q963: 28 W, 13 W idle
Q965: 28 W, 11 W idle
P965: 19 W, 10 W idle

975X: 13.5 W
955X: 13.5 W
945G: 22.2 W
945GZ: 22.2 W
945P: 15.2 W
945PL: 15.2 W

925XE: 13.3 W
925X: 12.3 W
915G: 16.3 W
915GV: 16.3 W
915GL: 16.3 W
910GL: 16.3 W

875P: 10.1 W
865G: 12.9 W
865GV: 12.6 W
865PE: 11.3 W
865P: 10.3 W

845GE: 6.3 W
845PE: 5.6 W

ICH:

ICH10: 4.5 W
ICH9: 4.0 W
ICH8: 4.1 W
ICH7: 3.3 W
ICH6: 3.8 W
ICH5: 2.4 W
ICH4: 2.2 W

I never would have guessed that after reading this:
http://anandtech.com/cpuchipsets/showdoc.aspx?i=2855

Moral of the story: Reality and theory don't always align.

The "reality" you're talking about is far from perfect to draw any conclusions from. That 975X board have an additional RAID controller which the P965 doesn't have. I'd like to see more mobos tested, but then again, the average 975X mobo have more components than the average P965.

You simply can't draw any useful conclusions between the power draw of P965 and 975X when the biggest difference in the test is 3.8 W.

BTW the P965 uses 10 W in idle in Windows XP.

If you add a graphics card, thus disabling on-board graphics,
would a G chip drop to the wattage of a P chipset? (e.g. G965->P965,
945G->945P)

Looks like the onboard graphics on the 965 chipset will consume a lot of power. I wonder if they can still be cooled passively.

Mats,

This is really good information. Where exactly did you get the numbers from tho? Did you dig into each and every tech doc for each and every chipset? Or is there some big comparison table? I looked briefly but the latter ones I found did not show power...

Yeah, I had to do it the hard way. Not that it was hard though, and certainly not one for each chipset. Some docs covers 4 or more chipsets, the doc titles says it all. I think it was 14 documents plus one for each ICH.

The TDP is found in every "Intel® xxx Chipset Family Thermal Mechanical Design Guidelines" doc, under the "Technical Documents" tab for each chipset. It is listed in the table of contents, and is usually found somewhere in pages 10 - 17. Here is the doc for the 965/963 series for instance (page 14).

28W is a lot to cool passively and if you look at the Asus P5B-VM which uses the G965, it doesn’t have a very large (passive) heatsink on the Northbridge. Not a good board for over-clocking I would surmise.
An extra 9W maximum for an IGP isn’t so much when you consider that an entry level GPU typically consumes 9W or more at idle; G965 has a TDP 9W more than the P965.

The tech docs also shows the power consumption for the 96x series in idle while running Windows XP. I'll add that now.

I think the TDP numbers for the newest chipsets will make people reconsider the choice of mobo, especially for HTPCs.

My guess is that it's just like with the nVidia 6100 chipsets, it's possible as long as you don't run games for instance.

I think you misunderstood me (my fault for being facetious in my post rather than elaborating on my line of thought). The intent of my comment is perfectly in alignment with your response - there's too many other factors involved to make an assumption about motherboard based on chipset (or vice versa).

In hindsight, I realize you originally posted the information to give guidance for chipset cooling solutions, not commenting on motherboards as a whole (which is what I was responding about)

pyogenes:Yeah, I realized that after a while. Not surprising, since English isn't my native language. It happens all the time.

Update: G31, G33, G35, P31, P35 and ICH9 added.

Hey, Mats, thanks for a really useful list for those of us trying to build a seriously low-power system. I was planning on replacing my p965 board for something cheaper to run, but maybe I'll hang onto it after all. 10W ain't so bad!

Of course you need to consider overclocking. Almost all ASUS motherboards overclock the MCH by default (in the BIOS it's called "Hyperpath 3"), so by default the north bridges consume more power than the Intel specs.

Remember that these numbers are more useful when discussing chip cooling rather than estimate mobo power consumption. Two different mobos with the same chipsets can have different power consumption. The power consumption also depends on MCH clock speed (like cmthomson mentioned), choice of components (and additional controllers like RAID), etc.

True, I hadn't thought of that. Does anyone know if turning off options in the BIOS also cuts of power to them? ie if I was to turn of my second SATA controller would it save me any power?

You have to understand how the mobo works, check this block diagram.
It probably will lower power consumption in the ICH, most likely not in the MCH since it won't transfer any data from the ICH to the CPU for that SATA channel since it's unused.

According to page 11 in this doc, the ICH uses 3.3 W when 4 untis (3 HD's in RAID and 1 optical drive) are connected,
and 3.7 W when 6 units (4 HD's in RAID and 2 optical drive) are connected.
Disabling devices in BIOS may affect boot time though.

Update: X38, G43, G45, P43, P45, X48 and ICH10 added.

I never knew that P45 would consume 6W's more than P35. I am glad i went for P35 board with downblowing cooler ^^. Excellent job Mats!!

Intel has a chipset for embedded systems or SFF systems that has a TDP of 2.3W and supports Atom – SCH US15W.
Panasonic has a Toughbook UMPC using the chipset with an Atom Z520 – Toughbook CF-U1 UMPC. 4.3W for the CPU + Chipset is starting to look good.

I did the exact same research, just didn't bother to post it!

Anyway, a couple oddball chips to add to the list:

Q35 13.5 W
3210 21.3 W
6702PXH 10.2 W

I was about to add the Q35 and the Q33, but the lower FSB made it a bit confusing, and probably the reason for lower power draw compared to G33.
I will probably add them later on.

Please add the 945GSE at 6 watts (+3.3 for the ICH7M)

http://ark.intel.com/chipset.aspx?familyID=35553

Total of 11.8W for the Intel Atom N270 + 945GSE and ICH7M (Acer Aspire One, EEEPC901+/Box, Wind, Wind PC)

Intel's power numbers are meaningless without their notes. The P35 numbers are higher than the G33 because of the power required to run a discrete card PCIe interface. The P35/G33/Q35/Q33 are all the same chip. If you use them the same way, they use the same amount of power. Intel even points this out:


So, all the 3-series chips are really 16 W TDP.

I vote no for including numbers from embedded low power chipsets like the 945GSE unless you can get those in retail motherboards. The 945GSE only has a single DDR2 channel and no PCIe x16 to power.

And why the 945GSE when all the other versions of the 945GM have dual channel DDR2 for 1W TDP more?

Q35 and X58 added.

One more that may be of interest: Q45: 17 W Max TDP, Idle (C1/C2): 6 W, Idle (C3/C4): 4.7 W.

I'm confused about the discrepancies between the P, Q, and G series (other than the P not having integrated video).



I sort of understand, but not really.



I added the Q45 line. Intuitively, the P series should have lower TDP in general due to lack of video. And looking at the P35 vs G35 case, this is true. Q35 throws it all for a loop though.

Now, why with P45 vs G45 is there no difference in idle TDP? And why did TDP go up from the P35 to P45? My understanding is that there weren't too many features added and there was a process shrink.

Anecdotal information: I believe the Q35 does in fact have fairly low idle power consumption. I built my fileserver using an Intel Q35 board with an E5200. I used my Kill-A-Watt, and let it run for over a week. The machine averaged about 118 W AC. Then I rebuilt it, changing only the motherboard and processor to the Gigabyte GA-MA74GM-S2 motherboard and 4850e. I'm at about five days (not quite a week yet), but I've averaged 117 W AC on this setup. Virtually the same.

Given that SPCR found the GA-MA74GM-S2 to have one of the best idle power draws, I'd say that the Q35+E5200 is practically an equal.

The integrated video circuits are still present in the P versions and still use current. Adding a external PCIe video card additionally loads the Northbridge. So, the TDP is actually higher for the G33/P35 with a video card than when using the IGP. G35 is not the same chip as G33/P35, it is more like the G45 which is a much more complicated IGP than the GMA 3100 in the G33/P35.

P45=G45 sort of a die shrunk G35
P35=G33=Q33=Q35 (=G31/P31?)

Why don't they disable the IGP circuits in the P-series chips?



It makes a lot more sense for me to look at it that way.

The Q45 still looks like the odd one out, though. I'm tempted to do another motherboard swap, and compare the Q45 to the Q35 and AMD 740G I've already tested.

X58: TDP 24.1 W, idle 8.5 W
Source, p 13.
US15W (new Atom chipset): TDP 2.3W, idle unspecified*
Source, p 423.
* With a Z530, it is said to idle at 5 W, as per a number of manufacturers.