A warning about Intel processor power consumption and high CPU usage

[Monkeh]

I think my point is pretty obvious. I don't mind using power but the concern is the watt per gigahertz of processing. I don't want to use 66% more power for a mere 24% gain in speed.

But how have you measured this?

um the manufacturers own published data ? they say 2.5GHz 15W which is what it is sold at and turbo mode of 3.1GHz 25W except that it will sit happily at nearly 3GHz non stop so not sure if it's meant to do that but windows allows it.

So you haven't measured anything.

TDP is a measure of the power a cooling system should be able to dissipate. It is not the current power consumption, the minimum power consumption, the maximum power consumption, or any other measure of actual power consumption at any point in time.

[senso]

I think my point is pretty obvious. I don't mind using power but the concern is the watt per gigahertz of processing. I don't want to use 66% more power for a mere 24% gain in speed.

But how have you measured this?

um the manufacturers own published data ? they say 2.5GHz 15W which is what it is sold at and turbo mode of 3.1GHz 25W except that it will sit happily at nearly 3GHz non stop so not sure if it's meant to do that but windows allows it.

No..
The 25Watts are only available IF the BIOS is setup to enable TDP-UP, if not, you will only ever use 15Watts..
Download HWMonitor, unzip it and watch the TDP values..

[Simon]

ok, I see well I can confirm that given the fan racket a substantially higher amount of power has been used at 3GHz than 2.5GHz.

[Someone]

ok, I see well I can confirm that given the fan racket a substantially higher amount of power has been used at 3GHz than 2.5GHz.

Yes, the fan noise might be annoying but you've not demonstrated the change in power consumption. There are many dumb fan control techniques out there that will suddenly switch on the fan for just a small increase in temperature.

[Simon]

ok, I see well I can confirm that given the fan racket a substantially higher amount of power has been used at 3GHz than 2.5GHz.

Yes, the fan noise might be annoying but you've not demonstrated the change in power consumption. There are many dumb fan control techniques out there that will suddenly switch on the fan for just a small increase in temperature.

So you say you are disbeleiving the the laws that tell us that a semiconductor device gets hotter when run faster and that when hotter it needs more voltage to run at the same speed ? More clock speed means more power consumption right there as the power is expended every time the gates switch so you will get at least a linear increase in power requirement, but then you need more voltage, clearly this is an exponential relationshitp, granted I do not know how exponential, but the fact that it shoots to 3GHz and then drops to 2.85GHz (oscilating from 2.84 to 2.86) tells me that it is creating more heat at 3GHz than the system was designed to tolerate for a long time and it has to reduce speed while maintaining some extra speed but clearly at increased power demand. Now I'm not exactly complaining, I just like to be aware that when I buy a machine that states 2.5GHz it won't try and kill itself if I want high usage for a longer period. In my case I have analysed the situation and because my use of the machine means long periods of high demand I have throttled it back whilst also only letting BOINC use 50% of processor time.

[PlainName]

    um the manufacturers own published data ? they say 2.5GHz 15W which is what it is sold at and turbo mode of 3.1GHz 25W except that it will sit happily at nearly 3GHz non stop so not sure if it's meant to do that but windows allows it.

So you haven't measured anything.

Might I suggest a download of Core Temp which will tell you exactly how much power your CPU is using in realtime, along with the CPU usage and details?

Not sure about Windows, but traditionally CPU usage percentage is the ratio of overall time vs time spent in the idle loop (that is, not doing anything in particular). Thus 100% usage would mean no time in the idle loop, and 0% would mean nothing happening except idle looping. For a dual-core CPU, an overall usage of 50% might suggest one core is compute bound, but bear in mind that the machine is doing other stuff too - OS maintenance, display, etc. So a completely bound core should show a figure of something over 50%.

Hence why Core Temp and the like is so useful - it will show you what each core is doing individually. If an overall 50% really is a completely bound core, Core Temp will show that as 100% for that core.

BTW, if you're running W7 (or 8 probably, or 10 with some third party app) Core Temp does a very nice desktop gadget.

[Someone]

ok, I see well I can confirm that given the fan racket a substantially higher amount of power has been used at 3GHz than 2.5GHz.

Yes, the fan noise might be annoying but you've not demonstrated the change in power consumption. There are many dumb fan control techniques out there that will suddenly switch on the fan for just a small increase in temperature.

So you say you are disbeleiving the the laws that tell us that a semiconductor device gets hotter when run faster and that when hotter it needs more voltage to run at the same speed?

You are exaggerating the change in power consumption, based on your arguments of the fan speed and your misinterpretation of the part specifications. Until you've found some way to measure the change in power consumption you're just waffling on against all of our real world experiences.

[Simon]

You are exaggerating the change in power consumption, based on your arguments of the fan speed and your misinterpretation of the part specifications. Until you've found some way to measure the change in power consumption you're just waffling on against all of our real world experiences.

So please post some of your real world experiences. Heavier power consumption at higher processor speeds is no mith, it's why the overclocking boys have to work so hard. I never got more than 5% if I was lucky out of stock gear. Obviously the processor was made to go up to 3.1GHz and use 25W but there is no specification on how long for. Laptop manufacturers have had to come clean on how long batteries really last and we have gone from laptops that lasted 5 hours in sleep mode to laptops that can actually be used for 3-4 hours. Clearly one way to achieve this is to make the processor efficient.

Now as I just said, we all know that running a processor faster at the same voltage uses more power. Are you or are you not aware that to run hotter at a higher speed requires more voltage (the basic overclockers tool). now can't you see that at the very least with a linear power increase with speed and an square law of power increase with voltage rise (basic ohms law) the speed/power relationship is going to be exponential. I am not saying by how much because as been pointed out I have not made measurements. Unfortunately the program suggested earlier won't diplay my power consumption. What I can say is that 3GHz 20% time usage = noisy fan with warm air coming out, 2.4GHz 50% time fan hardly moving and air is cool! now don't say you can't deduce anyrthing from that.

As to all the experiances and data "everyone else" has, please do share. If there is another program that can measure my processor power then please let me know what it is.

[KhronX]

Guys, i'll give you a hint:

https://www.techpowerup.com/downloads/2815/throttlestop-8-30

This will, on one hand, allow monitoring of multipliers / clock frequencies / momentary power consumption / temperatures, and on the other hand, allow access to power limit modification, enabling/disabling Turbo, EIST, power-saving and many other Intel CPU features.

[senso]

I said in page 2 for him to run hwinfo/hwmonitor, didnt care, better to ramble without any facts.

[KhronX]

Those are just for monitoring; ThrottleStop can also CONTROL some settings

[Simon]

I said in page 2 for him to run hwinfo/hwmonitor, didnt care, better to ramble without any facts.

Okay so you want some facts. Firstly if I can't control the speed of the processor how I might do comparative measurements. Hardware monitor as far as I know does not allow speed control. The tech power up program does. So what I am doing now is running BOINC on all four cores just to make sure the processor is fully loaded at 100% time use.

with turbo off (2.5 GHz) I get 11.4W with turbo on (2.86GHz) 15W, 14% speed increase 31.5% power increase! Did I or did I not say it was not a linear relationship ?

[KhronX]

https://en.wikipedia.org/wiki/CPU_power_dissipation#Sources

Technically, the increased voltage needed FOR that higher frequency, is what pushes the power higher, than the frequency itself.

That being said, if you're feeling adventurous, if you click on the "FIVR" button, that might give you some access to tweaking the voltage settings for various CPU subcomponents.

[Simon]

https://en.wikipedia.org/wiki/CPU_power_dissipation#Sources

Technically, the increased voltage needed FOR that higher frequency, is what pushes the power higher, than the frequency itself.

And didn't I say just that a couple of posts back. I said that there will probably be a linear increase in power dissipation with the frequency increase and due to ohms law the voltage increase will create a an exponential power increase probably to the square.

[KhronX]

Then use ThrottleStop to limit your max CPU clock (via setting the multiplier), and job done.

Otherwise, it's starting to sound like you're complaining about the laws of physics 

[Someone]

Running some real workloads I left the logging on and found that this i7 quad core actually gains efficiency as the cores are spun up. It seems the memory controller, graphics core, and static loads are high enough to mostly swamp the scaling effects. Actually I'm quite surprised that the compute/watt efficiency is higher as the additional cores are turned on. As always it pays to actually profile your use cases without jumping to conclusions, and it seems that the modern computer design paradigm of boosting clocks to terminate tasks as fast as possible is both better for the user and might use less power even on these bigger processors.

[Monkeh]

and it seems that the modern computer design paradigm of boosting clocks to terminate tasks as fast as possible is both better for the user and might use less power even on these bigger processors.

You mean the highly paid idiots at Intel actually did their research and didn't just bolt these features on because they sound good in the marketing blurb? That's unpossible!

[Simon]

Otherwise, it's starting to sound like you're complaining about the laws of physics 

No I'm just pointing the physics out to people. I never expected my CPU to boost for extensive periods and I am given no obvious control. What I am saying is that if people are running things like BOINC and care about power or fan noise then it pays to lock down the processor. I am please I have found the suggested program as it allows me to do controls on the fly.

[Someone]

and it seems that the modern computer design paradigm of boosting clocks to terminate tasks as fast as possible is both better for the user and might use less power even on these bigger processors.

You mean the highly paid idiots at Intel actually did their research and didn't just bolt these features on because they sound good in the marketing blurb? That's unpossible!

Well its not in their blurb anywhere I can see, they advertise the dynamic clocking/gating/power as a way to save power by reducing the processor power when its not needed but I havent seen these sorts of work/watt measurements discussed across a wide range of conditions. The traditional thoughts/analyses suggest the opposite I what I measured:
http://www.erudianart.com/research/papers/turboboost.pdf
Which is why its interesting.

[senso]

Otherwise, it's starting to sound like you're complaining about the laws of physics 

No I'm just pointing the physics out to people. I never expected my CPU to boost for extensive periods and I am given no obvious control. What I am saying is that if people are running things like BOINC and care about power or fan noise then it pays to lock down the processor. I am please I have found the suggested program as it allows me to do controls on the fly.

The CPU doesn't use 15Watts to do 2.something Ghz, my 4700 doesn't use 47Watts to iddle at sub Ghz speeds, nor does it use 90 or 120Watts to run at full tilt, open a monitor program and see -.-

Its a dynamic power usage that is always oscilating, I have hit 3.4Ghz with less than 25Watts..
And the fan does a racket because its undersized and the stock thermal paste is crap.
Repaste it with Artic MX-4 or Kryonaut and you will temp drops between 4 and 10ºC.

[Simon]

Yes my processor also shows speeds of 3 GHz whilst not using a lot of power but as soon as you actually put something onto it it slows down due to thermal throttling. The reality is my processor cannot use more than 15 W in power after that it starts to throttle. It might manage sixteen depending on the ambient temperature. The peak is eighteen, Intel claim it could use up to 25 W at 3.1 GHz but I've never seen actually get that fast.

It's a bit confusing because even when I had four threads going in BOINC the processor still only showed 50% loaded and windows also shows lowering load percentages when you reduce the processor speed.

[Towger]

TDP is not the power the chip is using, it is the amount of heat it generates.  The speed of fan in a modern system depends on a lot of factors.  It may not always ramp up linearly and as heat increases. The OS can play with it as well for power saving/performance mode etc.

[suicidaleggroll]

TDP is not the power the chip is using, it is the amount of heat it generates.

First law of thermodynamics.  It's not generating sound or light, it's not radiating a significant amount of power via EMI, it's not charging or discharging an energy storage device, etc., which means power in = heat out.

[Simon]

TDP is not the power the chip is using, it is the amount of heat it generates.

First law of thermodynamics.  It's not generating sound or light, it's not radiating a significant amount of power via EMI, it's not charging or discharging an energy storage device, etc., which means power in = heat out.

Don't worry, I work for a heat transfer company and someone started talking about the efficiency of a kettle but could not identify the losses.

[Zero999]

https://en.wikipedia.org/wiki/CPU_power_dissipation#Sources

Technically, the increased voltage needed FOR that higher frequency, is what pushes the power higher, than the frequency itself.

Actually just increasing the frequency on its own and keeping the voltage constant will increase the power consumption. This is assuming all things remain equal: the CPU isn't shutting down circuitry when it's not being used.