Interesting article on Nehalem power consumption

[VirtualLarry]

link

Power consumption of an IC is a function of the number of switching events and the square of the supply voltage, at least in theory. If leakage currents are taken into account, we found that the third power of the voltage provides a better fit for any of the CPUs we have measured over the years. What this amounts to is that a 10% increase in voltage will result in a 21% increase in power consumption and thermal dissipation using the classic square function and a 33% increase using our empirically derived function. If the operating frequency is increased by 30% (knowing that a lot of the Nehalem cores will do run up to 4.1 GHz) using 10% overvolting, the thermal dissipation will reach 177% of the stock value.

 

[OCGuy]

Originally posted by: VirtualLarry
link

Power consumption of an IC is a function of the number of switching events and the square of the supply voltage, at least in theory. If leakage currents are taken into account, we found that the third power of the voltage provides a better fit for any of the CPUs we have measured over the years. What this amounts to is that a 10% increase in voltage will result in a 21% increase in power consumption and thermal dissipation using the classic square function and a 33% increase using our empirically derived function. If the operating frequency is increased by 30% (knowing that a lot of the Nehalem cores will do run up to 4.1 GHz) using 10% overvolting, the thermal dissipation will reach 177% of the stock value.

Yes and what would the "Value" be of a 4.1ghz Nehalem if Intel released skus that high?


$3000?

 

[ilkhan]

the value being 130W, the stock rating for the bloomfields.

 

[Idontcare]

Originally posted by: VirtualLarry
If leakage currents are taken into account, we found that the third power of the voltage provides a better fit for any of the CPUs we have measured over the years.

What this amounts to is that a 10% increase in voltage will result in a 21% increase in power consumption and thermal dissipation using the classic square function and a 33% increase using our empirically derived function.

If the operating frequency is increased by 30% (knowing that a lot of the Nehalem cores will do run up to 4.1 GHz) using 10% overvolting, the thermal dissipation will reach 177% of the stock value.

My personal opinion is that these guys really should know better than to just put forth a modification to existing power-consumption equations without providing any data to backup such claims.

Want to make the squared-law suddenly a cubed-law so your claims on Nehalem power consumption sound all the more punchy? Then show us a graph with some data that clearly demonstrates a cubed-relationship between frequency and power-consumption.

Then make your punchy projections about Nehalem power consumption.

Then get a Nehalem and measure it at the wall and see what the relationship is between power-consumption and frequency in a dirty test like that.

 

[dmens]

power dissipation when overclocking will exceed the TDP? say it ain't so.

my biggest objection to that article is that there are no measurements whatsoever, and all power claims were made by multiplying the supply voltage with the recommended supply current for each voltage plane, which yields a number meaningless for any kind of discussion.

 

[pm]

Then get a Nehalem and measure it at the wall and see what the relationship is between power-consumption and frequency in a dirty test like that.

That was my thought too - why didn't they just measure it from the wall? It's hard to get a precise number for the CPU power in total, but it shouldn't be too hard to get relative numbers (the results will be a bit noisy, but just take the measurement multiple times and average).

 

[aigomorla]

the i7 is one of the first chips which can disable its unused core to go into power save.

So if you dont need the massive 8 threads, the cpu can shut off its cores, and conserve power.

Tubro on was made so users could get a bit more for there buck. Of course people who overclock, turbo on is only an excuse to get +1 on your multi, but to a non overclocker, intel has built in overclocking for them.

So if you throw in all the power save features the i7 has, it will draw less power then a yorkfield even.

Originally posted by: Ocguy31

Yes and what would the "Value" be of a 4.1ghz Nehalem if Intel released skus that high?


$3000?

http://i125.photobucket.com/al.../aigomorla/Final-1.jpg

my entire system? or just the board cpu and ram and cooling?

 

[pm]

aigomorla, those numbers are from air-cooled? What is the air temperature of the room you are in? Our house is sitting at about 19C most of the time... 18C minimum seems... low.

 

[Zstream]

Originally posted by: pm

Then get a Nehalem and measure it at the wall and see what the relationship is between power-consumption and frequency in a dirty test like that.

That was my thought too - why didn't they just measure it from the wall? It's hard to get a precise number for the CPU power in total, but it shouldn't be too hard to get relative numbers (the results will be a bit noisy, but just take the measurement multiple times and average).

That is insane to test at the wall. Talk about bad test, that is one of the worst. How do you know that the Motherboard, ram is not taking more current and thus skewing the results? How do you calculate the FSB voltage?

The claim was not an anti Intel bash for all the "Intel" people who are bashing the article.

 

[Concillian]

Originally posted by: pm
aigomorla, those numbers are from air-cooled? What is the air temperature of the room you are in? Our house is sitting at about 19C most of the time... 18C minimum seems... low.

Does Aigo have any air cooled PCs?

 

[aigomorla]

Originally posted by: Concillian

Originally posted by: pm
aigomorla, those numbers are from air-cooled? What is the air temperature of the room you are in? Our house is sitting at about 19C most of the time... 18C minimum seems... low.

Does Aigo have any air cooled PCs?

LOL is the correct answer.

 

[Idontcare]

Originally posted by: Zstream

Originally posted by: pm

Then get a Nehalem and measure it at the wall and see what the relationship is between power-consumption and frequency in a dirty test like that.

That was my thought too - why didn't they just measure it from the wall? It's hard to get a precise number for the CPU power in total, but it shouldn't be too hard to get relative numbers (the results will be a bit noisy, but just take the measurement multiple times and average).

That is insane to test at the wall. Talk about bad test, that is one of the worst. How do you know that the Motherboard, ram is not taking more current and thus skewing the results? How do you calculate the FSB voltage?

The claim was not an anti Intel bash for all the "Intel" people who are bashing the article.

What you do is you measure the power consumption at the wall for varying frequencies.

You make a plot of power-consumption versus frequency. It will not have a zero value for the y-intercept because the system power is non-zero but roughly constant in all the tests.

So what you should get is a set of data that conforms to a y = A*x^3 + C*x^2 + m*x + b.

What the lost circuits people are claiming is their historic data indicates the coefficient "A" is significantly larger than zero, so much so that it causes the term A*x^3 to dominate the remaining terms in the equation.

As for all the concerns you have regarding mobo, ram, PSU conversion efficiency, etc, those are captured in the m*x + b part of the equation (unless PSU efficiency is particularly crappy and not semi-linear in the amperage regime of interest here).

There is reason I said this would be a "dirty test"...what makes it dirty is that the tester must assume their PSU is not crappy, beyond that its a pretty straightforward test matrix.

 

[Jabbernyx]

Originally posted by: Idontcare
There is reason I said this would be a "dirty test"...what makes it dirty is that the tester must assume their PSU is not crappy, beyond that its a pretty straightforward test matrix.

Easy enough for enthusiasts like us to do and a meter like the Kill-a-Watt is accurate enough. Anyone actually run an OCing i7 setup on a blue-light special PSU?

 

[SickBeast]

Originally posted by: aigomorla
the i7 is one of the first chips which can disable its unused core to go into power save.

So if you dont need the massive 8 threads, the cpu can shut off its cores, and conserve power.

Tubro on was made so users could get a bit more for there buck. Of course people who overclock, turbo on is only an excuse to get +1 on your multi, but to a non overclocker, intel has built in overclocking for them.

So if you throw in all the power save features the i7 has, it will draw less power then a yorkfield even.

Originally posted by: Ocguy31

Yes and what would the "Value" be of a 4.1ghz Nehalem if Intel released skus that high?


$3000?

http://i125.photobucket.com/al.../aigomorla/Final-1.jpg

my entire system? or just the board cpu and ram and cooling?

aigomorla your system is stupid fast and your temps are amazing. Your i7 at 4.2ghz is cooler than my Opteron 165 was on the stock heatsink at 2.4ghz.

My guess is your rig is among the top-10 here at AT in terms of performance and temps. :beer:

You could probably make a living building extreme computers if you don't already do so. Just find some people with money to burn. They will want one.

 

[SickBeast]

Originally posted by: pm
aigomorla, those numbers are from air-cooled? What is the air temperature of the room you are in? Our house is sitting at about 19C most of the time... 18C minimum seems... low.

aigomorla is a watercooling guru.

I'm not allowed to talk about it, though. The last time it ended in an epic flame war.

Fight fire with water as they say. :beer:

 

[OCGuy]

Originally posted by: aigomorla
the i7 is one of the first chips which can disable its unused core to go into power save.

So if you dont need the massive 8 threads, the cpu can shut off its cores, and conserve power.

Tubro on was made so users could get a bit more for there buck. Of course people who overclock, turbo on is only an excuse to get +1 on your multi, but to a non overclocker, intel has built in overclocking for them.

So if you throw in all the power save features the i7 has, it will draw less power then a yorkfield even.

Originally posted by: Ocguy31

Yes and what would the "Value" be of a 4.1ghz Nehalem if Intel released skus that high?


$3000?

http://i125.photobucket.com/al.../aigomorla/Final-1.jpg

my entire system? or just the board cpu and ram and cooling?

Very nice. I cant wait until you hint at the new stepping ES for us

 

[zagood]

153x27 :Q

friggin unlocked multipliers.

 

[Idontcare]

Originally posted by: Jabbernyx

Originally posted by: Idontcare
There is reason I said this would be a "dirty test"...what makes it dirty is that the tester must assume their PSU is not crappy, beyond that its a pretty straightforward test matrix.

Easy enough for enthusiasts like us to do and a meter like the Kill-a-Watt is accurate enough. Anyone actually run an OCing i7 setup on a blue-light special PSU?

I'm generating the data as we speak, albeit on a QX6700 (changing the multi so I can keep everything else the same in the system) not with an i7.

 

[William Gaatjes]

Originally posted by: VirtualLarry
link

Power consumption of an IC is a function of the number of switching events and the square of the supply voltage, at least in theory. If leakage currents are taken into account, we found that the third power of the voltage provides a better fit for any of the CPUs we have measured over the years. What this amounts to is that a 10% increase in voltage will result in a 21% increase in power consumption and thermal dissipation using the classic square function and a 33% increase using our empirically derived function. If the operating frequency is increased by 30% (knowing that a lot of the Nehalem cores will do run up to 4.1 GHz) using 10% overvolting, the thermal dissipation will reach 177% of the stock value.

Is it not so, That leakage current's are primarily static currents ?
This means that these current's also flow when the clock signal is halted.

Nehalem as most consumer cpu's are build using cmos technology. These are transistors that only consume current during switching. That is because the gate(input) source(part of the output) capacitor of these (Metal Oxide Semiconductor Field Effect Transistors = mosfet's) has to be charged and discharged. This charging and discharging are the switch current's that flow. Now these current's are very small. nano amperes > microamperes for 1 transistor dpending on what it's function is. 1 nano ampere = 10^-9 , 1 microampere = 10^-6.
Now since we have billions of transistors, these currents have to be multiplied by 1000.000.000. to 1000.000 . That is where the power draw comes from. Now charging and discharging of capacitors with increasing frequency uses more current. See it as filling and draining a glass of water. The faster you do it, the more water you need to fill it up for the same amount of water in the glass. More water in a shorter time. Frequency is nothing more then 1 / (time you need to fill the glas). The shorter the time the higher the frequency. f = 1/t . Where f = frequency and t = time.

Basically the switching of the gate source capacitors is what draws power. And since the high k metal gate technology the static dc currents are very low for the cmoss proces of Intel. I doubt the nehamlem consumes that much more then other transistors. When compared to the old cores2 , it has an IMC, a power hungry qpi( with respect to the old gtl+ frontside bus) and more cache. In the end, nehalem does not does bad at all. If you want to compare nehalem's power usage, compare it too a similair core2 chip and add the chipset power to that core 2 power usage.

 

[Idontcare]

Here you go, Vcc versus power consumption at the wall for my system.

http://i272.photobucket.com/al...usPowerConsumption.gif

Note that the coefficients for both the squared and linear terms are practically zero (10^-9 and 10^-6). The y-intercepts represent system power consumption in the absence of the CPU (so chipset, ram, hdrives, vidcard, fans, PSU efficiency, etc) - comes in the vicinity of 105-120W which is reasonable for my components.

If I force the fit to be quadratic I get non-physical y-intercepts, 700W in one case and 14W in the other, neither of which could possibly be my system level power consumption so this counts as a second peice of evidence that quadratic power scaling for my system is not correct.

So in conclusion, for a 65nm Kentsfield quadcore chip, my results corroborate the lost circuits claims that power consumption scales as the cube of cpu voltage, not the square.

 

[pm]

That is a serious number of datapoints, IDC. You are amazing.

Interesting data too.

 

[Idontcare]

Originally posted by: pm
That is a serious number of datapoints, IDC. You are amazing.

Interesting data too.

It took waaaaay more time than I anticipated too because my areca card takes forever to initialize on reboots. It was rewarding though to see the end results.

I would have never believed my CPU exhibits x^3 power consumption as a function of Vcc unless I had seen it with my own eyes. I was amazed to see the results too.

 

[pm]

Ok. I tried something similar with my new Core i7 920.

All active components in the system:
Core i7 920 (133x20=2.67GHz), Asus P6T motherboard, 2x2GB Mushkin DDR3-1333 (1.5V), EVGA GeForce 512MB 8800GT (stock), Intel X-25M 80GB, Samsung Spinpoint 1TB, Lite-on DVD drive, Scythe Infinity cooler w/ fan, 3 fans, Antec Earthwatts 380W power supply

Measuring device is a La Cross "Cost Control" watt-meter. Measuring the 120V input to the Antec power supply. Testing methodology was to load Vista Ultimate 64, wait 15 minutes for everything to calm down, turn off non-essential services and processes, and I measured the peak value on the watt-meter while running the "XS Bench" benchmark in the Real Temp 3.0 program. I changed the voltage using Asus AISuite (from Windows - no rebooting, except for the crash at 0.95V). The wattmeter lists all values to the nearest integer. All measurements were taken 3 times, and there was only +/- 0.5W variation in any of the measurements.

0.95V -crash-
1.00V 129W
1.05V 131W
1.10V 132W
1.15V 134W
1.20V 136W
1.25V 137W
1.30V 140W
1.35V 142W

The motherboard bios is set to all defaults - except the CPU Vcore voltage. Load-line is enabled (it's default).



* I am not a spokesperson for Intel Corp. *

 

[Gillbot]

Originally posted by: SickBeast
aigomorla your system is stupid fast and your temps are amazing. Your i7 at 4.2ghz is cooler than my Opteron 165 was on the stock heatsink at 2.4ghz.

My guess is your rig is among the top-10 here at AT in terms of performance and temps. :beer:

You could probably make a living building extreme computers if you don't already do so. Just find some people with money to burn. They will want one.

Originally posted by: SickBeast

Originally posted by: pm
aigomorla, those numbers are from air-cooled? What is the air temperature of the room you are in? Our house is sitting at about 19C most of the time... 18C minimum seems... low.

aigomorla is a watercooling guru.

I'm not allowed to talk about it, though. The last time it ended in an epic flame war.

Fight fire with water as they say. :beer:

Please don't add to his already high soapbox. :laugh:

 

[Idontcare]

Originally posted by: pm
Ok. I tried something similar with my new Core i7 920.

All active components in the system:
Core i7 920 (133x20=2.67GHz), Asus P6T motherboard, 2x2GB Mushkin DDR3-1333 (1.5V), EVGA GeForce 512MB 8800GT (stock), Intel X-25M 80GB, Samsung Spinpoint 1TB, Lite-on DVD drive, Scythe Infinity cooler w/ fan, 3 fans, Antec Earthwatts 380W power supply

Measuring device is a La Cross "Cost Control" watt-meter. Measuring the 120V input to the Antec power supply. Testing methodology was to load Vista Ultimate 64, wait 15 minutes for everything to calm down, turn off non-essential services and processes, and I measured the peak value on the watt-meter while running the "XS Bench" benchmark in the Real Temp 3.0 program. I changed the voltage using Asus AISuite (from Windows - no rebooting, except for the crash at 0.95V). The wattmeter lists all values to the nearest integer. All measurements were taken 3 times, and there was only +/- 0.5W variation in any of the measurements.

0.95V -crash-
1.00V 129W
1.05V 131W
1.10V 132W
1.15V 134W
1.20V 136W
1.25V 137W
1.30V 140W
1.35V 142W

The motherboard bios is set to all defaults - except the CPU Vcore voltage. Load-line is enabled (it's default).



* I am not a spokesperson for Intel Corp. *

Unfortunately both the cubic and quadratic solutions result in equally plausible system power consumption so we can't rule either power equation out for your i7.

http://i272.photobucket.com/al...o_bucket/pm_i7_920.gif

What we'd need to do is generate an accompanying set of data on the same rig but with the i7 underclocked by about 20% (say around 2GHz). Such data would assist in eliminating one more unknown.