Replies to post #45339 on Intel Corp (INTC)

[wbmw]

Intel/AMD Power Consumption Analysis

Buggi, here's a bit of an analysis I did on power consumption. Basically, it's no more than determining the base power consumption (without CPU), and since it is the same system across multiple cores, we can then subtract that and find CPU idle and loaded power. I also calculate the % of loaded power each processor has relative to its rated TDP.

I took a guess at 60W for base system power, since we expect steppings like Conroe B-1 to dissipate about 20W in idle (according to the spec), and that is precisely what we see with the E6600. E6600 TDP also comes out to be 63W, which also agrees with this figure. I am happy with a number of other scores as well, and I certainly invite others to come up with ideas to refine this figure.

So using the 60W base figure and subtracting, this means the power figures for these CPUs are as follows (first, Celeron vs. Sempron):

 
 Processor      Freq    Cache   Idle   Load   TDP   % of TDP 
 Celeron 420   1.6GHz   512K    15W    19W    35W     54%    
 Celeron 430   1.8GHz   512K    16W    21W    35W     60% 
 Celeron 440   2.0GHz   512K    17W    23W    35W     66% 
   
 Sempron 3400  1.8GHz   256K     7W    49W    62W     79% 
 Sempron 3600  2.0GHz   256K     7W    52W    62W     84% 
 Sempron 3800  2.2GHz   256K     7W    56W    62W     90%

Note how much lower loaded power Intel's Celeron is compared to Sempron. I would guess that Intel has disabled power management for the Celeron, since idle powers are nearly as high as loaded power. Intel's TDP is now quite a bit more conservative (these could be 25W parts!), coming in well below AMD's TDP percentage. Let's look at Allendale cores vs. Brisbane.

 
 Processor       Freq    Cache   Idle   Load    TDP   % of TDP 
 Pentium E2140  1.6GHz    1M       9W    33W    65W     51%    
 Pentium E2160  1.8GHz    1M       9W    35W    65W     54% 
 Core 2  E4300  1.8GHz    2M      10W    36W    65W     55% 
 Core 2  E4400  2.0GHz    2M      11W    35W    65W     54% 
 Core 2  E4500  2.2GHz    2M      11W    37W    65W     57% 
   
 AX2     3600+  1.9GHz  2x512K     8W    59W    65W     91% 
 AX2     4000+  2.1GHz  2x512K     8W    62W    65W     95% 
 AX2     4400+  2.3GHz  2x512K     8W    67W    65W     103% 
 AX2     4800+  2.5GHz  2x512K     8W    70W    65W     108% 
 AX2     5000+  2.6GHz  2x512K     8W    75W    75W     100%

Now this is interesting. Allendale idle TDPs are quite a bit better than Intel's old Conroe TDP, and now quite competitive with AMD. But under load, it's even better. These could be 40W parts! But Intel calls them 65W. Brisbane, on the other hand, is really pushing the definition of TDP these days. These chips now dissipate quite a bit more under load. Now for some Conroe G0 stepping vs. Windsor F2/F3:

 
 Processor       Freq    Cache   Idle   Load    TDP   % of TDP 
 Core 2  E6550  2.33GHz    4M     15W    54W    65W     83%    
 Core 2  E6750  2.66GHz    4M     15W    58W    65W     89% 
 Core 2  E6850  3.00GHz    4M     15W    63W    65W     97% 
   
 AX2     5000+  2.6GHz    2x1M     8W    96W    89W     108% 
 AX2     FX-62  2.8GHz    2x1M    12W   134W   120W     112% 
 AX2     6000+  3.0GHz    2x1M    12W   138W   120W     115%

Amazing. Intel delivers competitive idle power, and absolutely creams the Windsor core in terms of power under load. In all fairness, today AMD has a more efficient 2.8GHz core with the 5600+, and this month they are supposedly delivering an efficient 6000+, but these numbers are well over TDP (as we've seen in many other reviews). It's like the old Smithfield Pentium D days, but today the shoe is on the other foot. The new G-step parts do get closer to TDP (relative to Allendale), but they never cross it.

I'd say that one thing this analysis teaches me is that Intel isn't just way ahead of AMD in terms of performance, but also way ahead in terms of power efficiency, and that AMD's ratings aren't always worth what they claim to be TDP. Your mileage may vary based on which review you read, but these tests certainly agree that Intel has come a long way, and the newer steppings of Core 2 micro-architecture have received an enormous benefit to power and performance.

Unfortunately for AMD, Brisbane is a dud, Barcelona is late, and Windsor is nothing more than brown bananas. I'm glad to have finally found a review that compared all these cores side by side, to get a better competitive landscape. I will certainly be bookmarking this post for future reference. I hope it was useful for people.