How power hungry was netburst clock for clock?

[MatTheMurdera]

Wouldnt the clock speeds be most of the power draw on the old netburst stuff? Has anyone done a clock per clock comparison between AMD and Intel on a power basis rather than on a performance basis? I ask because google doesnt seem to quite have what I was looking for...

 

[TechnologyCoordinator]

You can probably find the data you're looking for on AMD and Inte'ls websites. When all else fails, Wikipedia is a very strong source of data. You might have to do some calculations yourself.

What is your definition of clock for clock?

AMD wasn't too far behind Intel in power consumption on some models. AMD processors of course run at lower speeds while Intel's netburst processors topped out around 4ghz.

Good luck and post your findings.

 

[BaronMatrix]

Wouldnt the clock speeds be most of the power draw on the old netburst stuff? Has anyone done a clock per clock comparison between AMD and Intel on a power basis rather than on a performance basis? I ask because google doesnt seem to quite have what I was looking for...

Let's just say "You could have lit up Broadway,Off-Broadway, and Off-Off-Broadway with the amount of energy you wasted by not using an Opteron"

ACTUAL AMD billboard.

 

[Guest]

Intersting question. The thing is tnat it changed alll the time with the processor revision...

Im pretty sure my Northwood is comparable to a AMD.

I dont have any in formation but mayeb you can do this kind of math. Find a web site that gives a measured power Draw under full load for a given CPU, take that Draw and divide it by speed of the cpu in Mhz, then you'll have a What/mhz that you can compare.

If Iterations reads this he might have a better way to do it since he seams expert with Conroe charts

 

[BaronMatrix]

Most power tests measure form the wall and not from the socket. Load for NetBust was in some cases more than 100W greater than AMD.

 

[Guest]

Um right! did'nt think but I guess you still include Chipset and all that stuff inside the measurement, just make its harder to find 2 computer equiped witht he Same Hd's etc.

Most power tests measure form the wall and not from the socket. Load for NetBust was in some cases more than 100W greater than AMD.

Still his question was clock for clock.

So 100w greater would mean the intel take 1/3 more power cvlock for clock, most likely 1/5 or something. Maybe the relation is proportional(not linear thats for sur, maybe exponential) to the number a stages in the processor pipeline?

Intersting topic, not anywhere near the time to look into it =)

 

[ivan_lee05]

any link?



not all processors of netburst archi. are power hunger... the northwood core consume just like the athlon64 cores..

the northwood core has only 20-stage pipeline.. while the prescott has too long 31-stage pipeline.. due to its long stages of pipeline and higher clock speed(supposed to be 5Ghz above). prescott core reaches high TDP's.. a Pentium4 prescott core has 95w of TDP.

while a Pentium4 northwood core has only 52.5w of TDP.


a man speaking inside intel.. o_0

 

[BaronMatrix]

Um right! did'nt think but I guess you still include Chipset and all that stuff inside the measurement, just make its harder to find 2 computer equiped witht he Same Hd's etc.

Most power tests measure form the wall and not from the socket. Load for NetBust was in some cases more than 100W greater than AMD.

Still his question was clock for clock.

So 100w greater would mean the intel take 1/3 more power cvlock for clock, most likely 1/5 or something. Maybe the relation is proportional(not linear thats for sur, maybe exponential) to the number a stages in the processor pipeline?

Intersting topic, not anywhere near the time to look into it =)


he would have to check several sites for tests at different speeds. As I said, you can't easily measure just what the CPU is using.

 

[MatTheMurdera]

By clock per clock I mean say take an operon @ 2.0 GHz and compare its power draw to a presscot and northwood also @ 2.0GHz. Then maybe throw in C&C as well as the Intel version (i forget the name) to see how much they can save.

 

[MatTheMurdera]

You can Isolate it to just Mobo and RAM (even get rid of that with using AM2). Im sure it wouldnt be that much of a varible during power draw tests.

 

[TechnologyCoordinator]

By clock per clock I mean say take an operon @ 2.0 GHz and compare its power draw to a presscot and northwood also @ 2.0GHz. Then maybe throw in C&C as well as the Intel version (i forget the name) to see how much they can save.

It is very possible that an Intel Pentium netburst processor underclocked down to an AMD speed would use less energy, but it's performance would be terrible.

Remember, AMD chips have a extra units or something like that, that's why they can be Intel chips at lower clock speeds. Those "extra" parts use electricity.

 

[MatTheMurdera]

Of course Intel's extra cache might as well.

 

[BaronMatrix]

By clock per clock I mean say take an operon @ 2.0 GHz and compare its power draw to a presscot and northwood also @ 2.0GHz. Then maybe throw in C&C as well as the Intel version (i forget the name) to see how much they can save.

As I said since you can only mesure total power it's difficult to make a direct comparison. The two machines have to have basically the same parts. Since AMD has no "northbridge" it's hard to extrapolate what the actual difference in the CPU is.

 

[DrBlofeld]

Don't forget that AMD CPUs contain an onboard memory controller while Intel chips have the controller on the Northbridge. It may be difficult to make a fair comparison because of this unless you can isolate either controller or assume that the majority of energy used by the Northbridge comes from the memory controller.

 

[Ycon]

AMD processors consume more power per clock with the difference being not so great at lower clocked models and being greater at higher clocked models.

 

[MatTheMurdera]

could you please cite your reference?

 

[Multiplectic]

Could it be calculated from the TDP? :?:

 

[1Tanker]

Wouldnt the clock speeds be most of the power draw on the old netburst stuff? Has anyone done a clock per clock comparison between AMD and Intel on a power basis rather than on a performance basis? I ask because google doesnt seem to quite have what I was looking for...

Let's just say "You could have lit up Broadway,Off-Broadway, and Off-Off-Broadway with the amount of energy you wasted by not using an Opteron"

ACTUAL AMD billboard.All that education and job experience, and this is the best you can come up with? Go bore the Wall Street crowd, and leave the forums to people who actually care. :tongue:

 

[MatTheMurdera]

Yes, but you would need to have the same parts (minus RAM and Mobo, but they dont have huge power draw, and shouldnt make a difference if they have the same RAM speed and manufactuer), as well as finding out system draw minus processor.

 

[xsandman]

By clock per clock I mean say take an operon @ 2.0 GHz and compare its power draw to a presscot and northwood also @ 2.0GHz. Then maybe throw in C&C as well as the Intel version (i forget the name) to see how much they can save.

This doesn't make much sense because prescott needs to run at a higher clock in order to get the same performance. This is due to the larger pipeline. That is why AMD went away from using frequency to label processors. A 1.2GHz AMD had performance similar to a 1.5GHz Intel Northwood. This is why AMD advertised it as a 1500+, even though it was a 1.2GHz CPU. Therefore you need to compare performance/watt based off of benchmarks and power at the wall. This is the best way to compare performance, especially with conroe.

 

[1Tanker]

By clock per clock I mean say take an operon @ 2.0 GHz and compare its power draw to a presscot and northwood also @ 2.0GHz. Then maybe throw in C&C as well as the Intel version (i forget the name) to see how much they can save.

This doesn't make much sense because prescott needs to run at a higher clock in order to get the same performance. This is due to the larger pipeline. That is why AMD went away from using frequency to label processors. A 1.2GHz AMD had performance similar to a 1.5GHz Intel Northwood. This is why AMD advertised it as a 1500+, even though it was a 1.2GHz CPU. Therefore you need to compare performance/watt based off of benchmarks and power at the wall. This is the best way to compare performance, especially with conroe.You haven't been reading this thread. It's strictly about watts/GHz.....performance isn't a factor in this discussion.

 

[xsandman]

By clock per clock I mean say take an operon @ 2.0 GHz and compare its power draw to a presscot and northwood also @ 2.0GHz. Then maybe throw in C&C as well as the Intel version (i forget the name) to see how much they can save.

This doesn't make much sense because prescott needs to run at a higher clock in order to get the same performance. This is due to the larger pipeline. That is why AMD went away from using frequency to label processors. A 1.2GHz AMD had performance similar to a 1.5GHz Intel Northwood. This is why AMD advertised it as a 1500+, even though it was a 1.2GHz CPU. Therefore you need to compare performance/watt based off of benchmarks and power at the wall. This is the best way to compare performance, especially with conroe.

No, he is simply looking for an answer to the question -- if you pit the processors clock tick for clock tick, which consumes more energy?

I am assuming he wishes to compare things like CPU leakage, power draw, etc. under the same conditions physically and not by trying to match up performance.

You are quite correct, the long pipeline requires extra power and is less efficient IPC (instructions per second wise), thus the need to ramp high clocks to achieve similar computational performance.

We can estimate it though without going into great detail. The formulation for dynamic power in a processor can be calculated by:

Power = C * V* V * F (C is cumulative capacitance at the gate level, V is drive voltage, and F is frequency).

Thus if you were to scale a 3.6 GHz P4 with a dissipation of 105 watts from 3.6 to say FX-62 speed which is 2.8, then the power should scale roughly about 2.8/3.6 or 0.77 which would put power at 81.7 Watts. Lower than an FX-62 of 125 watts, and even lower than a 4800+ at 89 watts.

This rather simple and straight forward analysis, though, is not entirely correct --- it is back of the envelop pure and simple. Don't take it as absolute, the analysis will require much further digging and explanation.

Jack

While i understand what he is going after, i am just saying that this analysis has little purpose since the cores are so different. Comparing frequency/power accomplishes little because every part of the pipeline is smaller on a P4 CPU. With a transistor of a constant speed will consume X amount of power. If a logic device is Y transistors deep, then it will consume XY Watts/clock. If a different logic device is twice as deep, it will consume 2XY Watts/clock, but will run at 1/2 of the speed but do approximately the same amount of work. (based off of simple logic, not actual power equations).

 

[Ycon]

 

[fainis]

what are you looking for my friend is called TDP .....just look on intel site ..and select your netburst processor

 

[1Tanker]

what are you looking for my friend is called TDP .....just look on intel site ..and select your netburst processor

That's not going to help compare Intel with AMD at same clockspeed. AMD rates their chips TDP differently.