actual power tested: Core 2 Duo / Netburst vs Athlon

[jojo4u]

German webpage hartecs4u.com measured the actual power consumption by measuring the current from the VRM to the CPU.
Example picture of setup here, here, here and here.

Burn-in tools for dual-core: 2x Prime95/1x SuperPI/BurnMax. The usage changed over time, you can see the used tools in the header of the diagrams.

Test equipment sounds and looks quite professional. (link)

C1E and EIST where enabled, I asked the author about that in the forum. For AMD you have both CnQ on/off.

AM2 Athlon X2 including 45 W 65 nm parts.


Intel Conroe including L2-stepping


Core 2 Duo and Netburst


AMD results:

Older test, no CnQ, BurnMax used (link):

Even older test, Cnq on/off, Prime95 (link):
link

edit: one link corrected, one added, newer tests

[cAPSLOCK]

You posted the AMD graph instead of the Intel one. [edit: either I messed up, or you made a correction ]

That's a pretty crazy mod of those guys, but it makes cross-platform comparisons possible. Pity they didn't have the conroes with the smaller cache, it would be interesting to see if there was a difference in idle consumption, the two they had there seemed to have pretty much the same idle consumption. This sort of confirms what I noticed in the article about the energy efficient athlons: the AMD's seem to use less when idling (I wasn't looking at the FX), although the difference is pretty small here, they do have a memory controller on board, and these are the normal athlons. Pity they didn't have speedstep, it would be interesting to see how much difference it makes.

My personal conclusion: AMD X2's are the better choice for average joe's, not only because you don't really need the power of a conroe, but also because they use less electricity at idle.

[croddie]

Which results are Core Duos may I ask? The AMD processors have their familiar names but what do the serial numbers for the Intels represent?

[jharne]

I'm sure some would have said at one point, AMD viral marketers probably, but it's not a fair test. AMD and intel chips have diverged so much, total system power is the best bet.

No wait someone already did.

[McBanjo]

Which results are Core Duos may I ask? The AMD processors have their familiar names but what do the serial numbers for the Intels represent?

E6800 and below

[jaganath]

I'm sure some would have said at one point, AMD viral marketers probably, but it's not a fair test. AMD and intel chips have diverged so much, total system power is the best bet.

No wait someone already did.

Huh? Are you trying to say CPU power consumption is irrelevant?

[vitaminc]

I'm sure some would have said at one point, AMD viral marketers probably, but it's not a fair test. AMD and intel chips have diverged so much, total system power is the best bet.

No wait someone already did.

Huh? Are you trying to say CPU power consumption is irrelevant?

CPU power consumption only matters to HSF designers. And it's not all that important compare to total system power consmuption at the moment, especially when your gfx card or northbridge could draw more power at Idle than your CPU.

What's important is the total power draw at the wall. That's going to effect your monthly electricity bill.

[cAPSLOCK]

What's important is the total power draw at the wall. That's going to effect your monthly electricity bill.

Granted, but that doesn't mean it's uninteresting to see power consumption of individual components, since most of us here buy the components separately.

[Mats]

No A64 revision F2, too bad. While I don't think they're so much different from the previous ones, it's still important to remember that the CPU's with low TDP probably becomes "EE" CPU's and are sold at a higher price.

No C2 with stepping B2, which is the production stepping. I have really no idea if they are any different from the B1 when it comes to power consumption though.

[Mats]

jojo4u: Could you please ask the reviewer about the TDP for the A64 revision E CPU's? Are those TDP's in the review individual ones or just taken from a list? Since they could be anything between something like 40 and 100 W it's pretty important.

I just saw their comment about AM2 mobos. Keep us updated if/when they test AM2. Thanks!

[aidanjm2004]

Do the Core 2 Duo cpus or their chipsets have a Cool n Quiet or Speedstep style feature to reduce power consumption at idle?

Seems like an AMd X2 would be a more environmentally friendly (as in, use less energy) choice for a jukebox or htpc if it is going to be left on 24 hours a day (as mine are)..?

[keyne]

C2D does have EIST (speedstep).
It also has alot of undervolting potential, i'm running an E6400 at 1.00V core. the CPU also has features to turn off parts of it during idle. Mine runs very cool...
Don't have equipment to measure power consumption.

[vitaminc]

What's important is the total power draw at the wall. That's going to effect your monthly electricity bill.

Granted, but that doesn't mean it's uninteresting to see power consumption of individual components, since most of us here buy the components separately.

Completely disagree.

People have choices between platforms but not individual components. For example, you can get a 9W Yonah, S939 NF4 motherboard, 256MB of DDR, but they wont work together.

I think its very misleading to compare CPU to CPU instead of platform to platform, and that's exactly the same rationel presented in AMD's Tech Analyst Day.

Further, it would be very interesting to see EIST (speedstep) disabled idle power consumption for the C2D. I am suspecting EIST is not enabled.

The test also need to include the system setup when CnQ and EIST kicks in. Listing a 3800+ (2.4GHz) with 4.64W power draw is misleading as it is probably running at 0.9V, 100MHz HTT x 5 mode.

[Mats]

I think its very misleading to compare CPU to CPU instead of platform to platform, and that's exactly the same rationel presented in AMD's Tech Analyst Day.

Sure, but it's still interesting from a CPU cooling point of view.

Further, it would be very interesting to see EIST (speedstep) disabled idle power consumption for the C2D. I am suspecting EIST is not enabled.

Me too. I get this feeling that it doesn't work properly yet, or at least not in some cases. Xbitlabs review is only one example, with 26 W in idle no matter if EIST is enabled or not.

[Mikael]

The test also need to include the system setup when CnQ and EIST kicks in. Listing a 3800+ (2.4GHz) with 4.64W power draw is misleading as it is probably running at 0.9V, 100MHz HTT x 5 mode.

They give you the idle figures for both C'n'Q enabled and disabled (4.64W and 8.16W respectively). What more do you want?

[paapaa]

Listing a 3800+ (2.4GHz) with 4.64W power draw is misleading as it is probably running at 0.9V, 100MHz HTT x 5 mode.

Why is it misleading? When a processor is doing nothing it is quite clever for it to undervold and underclock itself. I'm surprised why C2D has so "high" idle power consumption. It could most likely manage with less power than AMD processors if properly designed.

[smilingcrow]

I think its very misleading to compare CPU to CPU instead of platform to platform, and that's exactly the same rationel presented in AMD's Tech Analyst Day.

Sure, but it's still interesting from a CPU cooling point of view.

Further, it would be very interesting to see EIST (speedstep) disabled idle power consumption for the C2D. I am suspecting EIST is not enabled.

Me too. I get this feeling that it doesn't work properly yet, or at least not in some cases. Xbitlabs review is only one example, with 26 W in idle no matter if EIST is enabled or not.

From the perspective of quiet computing AND energy efficiency, why not try this:

Benchmark a system for performance whilst monitoring power draw (idle & load) at the maximum CPU speed that you can cool at a safe temperature limit for the CPU whilst keeping the noise level below a certain threshold.

To do this it makes sense to standardise on a case, power supply, heatsink, fan setup and a noise floor. I would personally choose a P180, Ninja+, Nexus 120mm fans (2 or 3); or similar. I also suggest using a passive power supply to take its noise out of the equation.
You could test with fan speeds at a number of settings in the range ~ 5 – 12V.

Choosing a safe temp for a CPU is a bit more subjective, particularly when comparing a mobile CPU to a desktop chip, as the former are typically rated at much higher temps.
Choosing a useful benchmark suite is also problematic. Utilities such as CPUBurn seem too abstract and don’t even use the SSE units that many (most!) high performance applications will use. It seems likely to me that using SSE will be more energy efficient than not doing so, and since both Intel & AMD use these instructions there isn’t a good reason not to focus on this.

I don’t think benchmarking on an open bench is helpful as it doesn’t give a real world environment for people to compare to. Again, it’s too abstract.
In my experience it’s possible to run a C2D at 2.9GHz in very close to silence. This level of performance is going to be high enough for all but the most demanding user, which is why I agree that CPU power consumption isn’t such an important issue today.
When AMD release 65nm chips, it will be interesting to see how the AM2 platform compares to the C2D platform for power efficiency. By then both the AM2 & C2D platforms will be more mature and AMD will be more competitive with Intel on the CPU level, so it seems like a good time for a comparison.

Note: SSE allows both Integer and Floating Point calculations to be run in parallel. I developed signal processing applications that utilised SSE (Version 1) back in the late 90s and it was so easy to use SSE that I can’t imagine that many companies don’t use it. Intel provides excellent libraries that support SSE which used to be free but now cost a minimal amount.

[Poodle]

Ok so what's the score here, what's next?

I've been tempted by the C2D. But worried about Eist. It doesn't really seem to run as good as Cnq, does it?

That C2d is best at load, no question about that but idle is more important.


As we are starting to talk about total system power. I have two systems in mind right now:

The M2R32-MVP and the low power AtiRd580/sb600 chipset with a Tdp of like 8w and 2*16x pci-e from asus and a X2 3800+

http://www.asus.com/products4.aspx?mode ... =101&l3=09

Or the expensive not yet here Dfi with rd600/sb600 that will probably be more than 8w becuase of the memctrl. I will use a E6300, probably. The Nb has somewhat bigger cooler as you can see which makes me believe that it's a bit hotter than the rd580.

Message from Mod: Please embed long links with the url tag.

[croddie]

Some power tests here seem to suggest the idle power of core 2 duo is competitive with AMD's stuff:
http://techreport.com/reviews/2006q3/co ... ex.x?pg=16
Not sure what would account for the difference between tests.

[jojo4u]

Some power tests here seem to suggest the idle power of core 2 duo is competitive with AMD's stuff:
http://techreport.com/reviews/2006q3/co ... ex.x?pg=16

From the test:
"You'll notice that the results below include numbers for the Energy Efficient versions of the Athlon 64 X2 3800+ and 4600+. AMD sent these CPUs out to us along with a more power-efficient motherboard than our Asus M2N32-SLI Deluxe test platform, whose nForce 590 SLI chipset seems to be something of a power hog. The board AMD sent, however, is not an enthusiast-class mobo with dual graphics slots, so we elected not to include it in our tests. We wanted to test the EE chips opposite the Core 2 Duo on an enthusiast-class board, so we stuck with the M2N32-SLI Deluxe. It's possible that enthusiast-class boards based on the Radeon Xpress 3200 or the nForce 570 SLI chipsets could lower power consumption for all of the Athlon 64 processors here without compromising performance."

Core 2 board: Intel D975XBX - Intel 975X

[jojo4u]

jojo4u: Could you please ask the reviewer about the TDP for the A64 revision E CPU's? Are those TDP's in the review individual ones or just taken from a list?

Sorry for the late answer, but better than none ...

The numbers are perfectly fitting the plain TDP from a list. No word about individual ones. They also say all CPUs are from early production batches, so power consumption may be lower from more mature batches.

[jojo4u]

One thing bugs me about the Conroes. The 2 MiB cache version 6300/6400 seem to need more power than the 4 MiB version.

source 1:
http://www.lostcircuits.com/cpu/low_e/5.shtml
[quote="http://www.lostcircuits.com/cpu/low_e/2.shtml"]
In the case of the Intel Core2 Duo E6300 we need to stress the disclaimer that the “half-cacheâ€

[jojo4u]

http://praxis.thgweb.de/2006/12/07/stro ... age13.html (12V CPU rail tested)

There is a pattern developing ... Let's hope the announced L2 stepping fixes it.

[jojo4u]

Some evidence against:
http://www.hardtecs4u.com/reviews/2006/ ... index4.php
Updated the first picture in the first post.

Also anandtech notes that the vcore of their 6300/6400 is higher:
http://www.anandtech.com/cpuchipsets/sh ... i=2893&p=2
See next page for the result at idle. Rated vcore should not affect idle power consumption because of C1E/EIST but could indicate the quality of the circuits.

[jojo4u]

Xbitlabs has posted Athlon 64 90nm/90nm+EE/65nm and Core 2 Duo results only including the VRMs.

http://www.xbitlabs.com/articles/cpu/di ... ne_10.html

[jojo4u]

First post updated for E6420 and E4300. Notice the low power consumption! Also interesting that the E6420 has still revision/stepping B2.

[line]

I hope X-Bit can come up with a roundup of their own to confirm these numbers. Something along the lines of "power consumption and noise level of contemporary CPUs", similar to the GPU roundup they put together last year.

[jojo4u]

Two new diagrams including Athlon X2 and Core 2 Duo L2-stepping.

One thing to notice: They tested on the Asus M2N-E (nforce 570 ultra) and the Asrock Alive SATA2-GLAN (K8T890). The Asrock was significantly worse under load than the Asus.
The Asrock seems to have a cheap 4-phase VRM while the Asus uses a better designed 3-phase VRM. The article suggests that the more phases, the worse the efficiency. Also they say that the Asrock may have worse cooling for the VRM since the delta 6000+ is greater. They also noted that the retention module of the Asus didn't seem to have as much pressure, leading to a worse cooling of the CPU and thus to higher power consumption.

[NeilBlanchard]

Hello,

Two new diagrams including Athlon X2 and Core 2 Duo L2-stepping.

You mention the Core 2 Duo, but I don't see them on the graphs?

[jojo4u]

Sorry, forgot to mention that I updated the first post of this thread.