45nm Intel Wolfdale-Yorkfield and real power consumption

Maverickâ„¢ · Post by **Maverickâ„¢** » Sun Apr 13, 2008 3:25 pm

Hi guys, these days I'm thinking to do a new system for crunching
so I'm trying to understand which is the real power consumtion of these processors
there's a bit of confusion regarding this aspect because of the differences in the figures measured by some specialized sites

Xbitlabs

QX9659/9770 by Tom's

QX9650 by Anandtech

what do you think?

croddie · Post by **croddie** » Sun Apr 13, 2008 4:32 pm

Edit: sorry I misread

yuu · Post by **yuu** » Sun Apr 13, 2008 5:36 pm

how can u tell? tom is also cpu only

there is mofset efficiency and cpu voltages differ there is the difference.

Blue_Sky · Post by **Blue_Sky** » Sun Apr 13, 2008 7:07 pm

Those numbers are interesting.
I really don't feel like digging through those sites to find the relevant articles, but unless I am missing something important, those numbers don't really complement each other.
When in doubt though, stay far away from Tom's. They are the odd man out when you compare numbers, way too often.
Another thing to remember is that every site has a different methodology - numbers that should be similar vary by a lot due to this.

I've heard that processor power draw increases proportional to clock speed squared quoted a number of times here, which the Anandtech graph seems to illustrate nicely.

Are you trying to get the most bang for your watt or do you want it done as quick as possible?

mattthemuppet · Post by **mattthemuppet** » Sun Apr 13, 2008 9:25 pm

Tom's and Anand's figures are pretty similar for the QX9650 (quad core) and the Xbitlabs are for dualcores, showing that the 45nm CPUs (E8***) are more efficient.

All depends on whether your software can make use of >2 cores. If it can, then one of the new 45nm quads (Q9***) would be perfect.

ZircularLogic · Post by **ZircularLogic** » Mon Apr 14, 2008 5:19 pm

The xbitlabs numbers for the wolfdales are in line with what I calculated for my own system. (They are a touch high but they are probably using higher voltage than I do, which is the absolute minimum necessary to be stable.)

smilingcrow · Post by **smilingcrow** » Tue Apr 15, 2008 1:54 pm

Itâ€™s not rocket science, if your application supports 1 or 2 cores then buy a 45nm Core 2 Duo, if it supports more than 2 cores buy a 45nm Core 2 quad. Which one you buy depends on your budget and preference for over-clocking. Kirk out.

yuu · Post by **yuu** » Fri Apr 18, 2008 2:37 am

xbitlabs have the vrm loses exluded somehow

HammerSandwich · Post by **HammerSandwich** » Fri Apr 18, 2008 5:36 am

Blue_Sky wrote:I've heard that processor power draw increases proportional to clock speed squared quoted a number of times here...

No you have not! At least I hope not.

CPU power draw is roughly proportional to:
* frequency
* voltage squared
The reason we see huge power increases in the above results is that the testers must increase Vcore to hit big OCs.

This has been supported by my more limited testing of an E3110. Let me give you some examples. Running Orthos, I noted AC power draw and the BIOS Vcore setting. 450x7 = 3.15GHz is stable at 1.225v and 104W. 450x8 = 3.6GHz uses the same voltage and pulls 109W. The CPU needs 1.345v to hit 450x9 = 4.05GHz, and power jumps up to (sit down!) 156W. I've decided to run at 3.6...

fwki · Post by **fwki** » Fri Apr 18, 2008 6:39 am

Voltage is the key, not only for OC'd power use but at stock speeds as well. I found the Q9450 to be a tremendous undervolter. At Prime95 4-core load with 1.225 VID, my system pulls 153 watts and with 0.99375 VID, it pulls 115 watts, both stable. That's a 38 watt drop from the cpu alone.

AZBrandon · Post by **AZBrandon** » Fri Apr 18, 2008 7:48 am

fwki wrote:Voltage is the key, not only for OC'd power use but at stock speeds as well. I found the Q9450 to be a tremendous undervolter. At Prime95 4-core load with 1.225 VID, my system pulls 153 watts and with 0.99375 VID, it pulls 115 watts, both stable. That's a 38 watt drop from the cpu alone.

What clock speed are you running it at? That is a huge undervolt, wow! I've been considering upgrading my hardware to a quad core and looking between the Q9300 and Q9450. The Q9450 definitely costs more for little to no more clockspeed, but it has double the cache, which may be a big deal, especially for folding, for example.

Anyway, on the topic of folding, my main complaint has always been with total system power drawn. I'm not a big fan of spending an extra 80 watts to fold, but if you're pulling just 115W total system, I'm guessing that's only 20-30 watts over idle or so, which is way better.

amyhughes · Post by **amyhughes** » Fri Apr 18, 2008 9:13 am

Does this mean that if I take a 65W processor like an E8400 and clock it at half its rated speed and also undervolt it I can expect mobile-like power usage?

fwki · Post by **fwki** » Fri Apr 18, 2008 10:00 am

AZBrandon wrote:What clock speed are you running it at?

Stock speed for the Q9450 is 2.66GHz (8x333) and under Prime95 it's pegged at 8x and 115 watts. At idle it drops to 6x333 and the idle system power drops to 77 watts with 0.99375 VID. When manually set, voltage stays at 0.99375 VID under all multipliers. The system is very stable gaming, media encoding and under stress test.

See this thread for more details: viewtopic.php?t=47335&highlight=

fwki · Post by **fwki** » Fri Apr 18, 2008 10:19 am

amyhughes wrote:Does this mean that if I take a 65W processor like an E8400 and clock it at half its rated speed and also undervolt it I can expect mobile-like power usage?

Right now I am kicking myself for spending way too much on my home server mobile-on-desktop setup (see sig). I think these new 45nm will kill what little market there was for MODT. As far as E8400 (9x333FSB) goes, it probably doesn't make sense to underclock it when Intel will do it for you and put $100 in your pocket with the E3140 (8x266FSB).

Nordic Hardware: "Intel is planning on replacing the current 65nm Core 2 Duo 2000 series with a new 45nm series series called Core 2 Duo 3000. The series will sport 1066MHz FSB, 1.5MB L2 cache (a quarter of 6MB) and the first model of the series will be Core 2 Duo E3140, which operates at 2.13GHz. Price will be $75 in units of a thousand. The first shipments should arrive in June. The high multiplier and attractive price will most likely make it a very attractive toy for benchers and people looking for the most bang for the buck."

If these things undervolt like the other 45's (and the E2000's), they'll be putting them in laptops!

Plekto · Post by **Plekto** » Fri Apr 18, 2008 10:41 am

Looks lovely.

Expect it to generate PDA levels of heat.

Esben · Post by **Esben** » Sun Apr 20, 2008 3:34 am

My own measurements and calculations seem to fit excellently with the numbers of xbitlabs. The numbers from Anandtech back them up, considering their CPU has twice as many cores = twice the power consumption.

xafier · Post by **xafier** » Sun Apr 20, 2008 5:37 am

I have an E8200, I've overclocked it to 3.2Ghz, its running on 1.1125v instead of stock voltage of 1.225v... from what I worked out its actually using less power overclocked than it would stock with stock voltage!

The 45nm chips are great, you can get a healthy balance of power and power usage with these things