A warning about Intel processor power consumption and high CPU usage

[Simon]

I've recently made an interesting discovery. As far as I know this only applies to Intel processors as this is the manufacturer of processor I have experienced it on. Specifically i7-6500U in my new laptop. Despite not being a fan of the I series when I came to replace my laptop I found that the only machine that all round had the specification I wanted was in fact one with an i7 processor. So the first thing I set about about was trying to work out how to get programs to use specific processor cores. Core 0 is a physical core. Core one is it's virtual core so all it actually allows you to do is queue up more work there is no real processor behind it to do any work. The same goes for core two and three.

I quite quickly found a program called process lasso which claims to help schedule tasks to make more efficient use of the processor. Personally I couldn’t say one way or the other but fact is it does allow you to assign specific programs to specific physical cores.

However what I have noticed is that although my processor is rated for 2.5 GHz it will run up to 3.1 GHz although often closer to 3 GHz. But after doing some investigation I discovered that the power rating at 2.5 GHz is 15 W whilst for a mere increase of 600 MHz the power consumption jumped right up to 25 W. Another problem for me is that I run BOINC which makes constant use of the processor. This means that the processor is always forced into its over clock frequency which it seems it can do non-stop if necessary. Obviously given the drastic jump in power consumption this means an awful lot more heat as well as an awful lot more power being used and the fan makes one hell of a racket. But there is a solution. If you go into to the advanced power options of Windows you can limit the percentage of processor speed. This is a bit odd because I have it set at something like 95% but that limits the speed to 2.6 GHz. I think there is a little bit of confusion around what the percentage relates to in terms of the normal clock speeds and the overclocked speed.

The problem I had was that I was running BOINC at 20% time usage in order to try and control the noise as often I go to sleep with the laptop on. But once I had limited the maximum processor speed to something like what it was meant to do I actually found that for some reason the programme still never uses more than 2.4 GHz even though the new rating is 2.6 GHz. This is something I have found throughout Intel processors it’s a bit odd what speed it actually goes to even though maximum usage is being demanded. But I can now run BOINC at 100% at 2.4 GHz (which it chooses to do because the processor is allowed to go up to 2.6 GHz) and the fan makes no more noise than it did when it was running on 20% time at 3 GHz. Obviously I’m also using a hell of a lot less power which is a nice thing as well because the laptop as I just said often gets left on overnight. Not much of a problem in the winter as it contributes to the heating of the room anyway but I find it rather odd that the 25 W figure for maximum overclocked speed is hard to find on the Internet. It is repeatedly claimed to be a 15 W processor. Obviously the extra 600 MHz are very welcome but the power penalty for me does not justify it unless for occasional spurts. Say if I was auto routing a circuit board I would be happy for it to run to that power for a few minutes in order to get the routing done more quickly but not continuously.

Of course the thing to remember with these Intel processors is that the virtual core is not actually a processing core it is just extra instruction and data queueing circuitry that while it will slightly increase the performance of the processor at the end of the day once you have hit 50% usage you are in effect at 100%. Perhaps given the extra efficiencies the virtual cores bring about you could maybe stretch it to 60%. I do in fact have BOINC set to use only two cores for this reason, using more than two cores would just cause a conflict and on my work computer that has an i7 this actually caused absolute mayhem and tasks were failing to complete.

[IanB]

I would say that for most normal computing needs you should not worry about cores and core allocation at all. Just let Windows do it's own thing with task scheduling and don't try to intervene.

I believe BOINC should be written so it only runs in the background when the system is otherwise idle. Does it have a control panel that limits what fraction of the machine it can use? Maybe set it to 10% or something?

The advanced power option to limit the fraction of processor speed is a good one to know about. I think many people are not aware of it.

[Simon]

BOINC can be limited. You can tell it how many processor cores to use and how much time it is to use processor for but that is a bit of a blunt instrument. There is no way of getting it to use a maximum speed though. Yes on the whole allocating programs to processors is not a problem but with the I type of Intel processor if you have a heavy task that is multitasking or you have something like BOINC that will put more than one job on until it fills all cores none of these programs can distinguish between real and virtual cores so mayhem ensues. The thing that I'm not keen on is that there is no simple software switch to disable automatic overclocking.

BOINC can also be set to run when you are not using the machine in conjunction with the above limitations. Personally I don't find it causing a problem to the machine while I am using it. The only exception is on and I seven processor because you end up with two programs trying to use the same physical processor and sometimes your own program does not get enough responsiveness. I found this with 3D CAD programs at work that's become very jerky on the display when rotating models if all eight cores of the machine are under 100% use because this means that eight programs are trying to use 100% of four cores.

[shteii01]

Yeah, the whole i series of intel cpu is like runaway train.  They left me behind when they moved past i5 2xxx series.
Personally I am happy to see that used i5 2xxx series machines (complete machines, just add hdd/sdd) are hitting 100-150 dollar mark.  I might buy a couple of the little ones and set them up for separate tasks, like one machine for cad, one for programming uC, one to play games.

[Simon]

I don't know what the difference between the i3, i5 and i7 are, but I suspect the power profile is the same on all as it is common across processors, more speed = more heat, so for more speed more voltage is required.

[nctnico]

Yeah, the whole i series of intel cpu is like runaway train.  They left me behind when they moved past i5 2xxx series.
Personally I am happy to see that used i5 2xxx series machines (complete machines, just add hdd/sdd) are hitting 100-150 dollar mark.  I might buy a couple of the little ones and set them up for separate tasks, like one machine for cad, one for programming uC, one to play games.

That wastes a lot of energy! Better get one beefy PC and use virtual machines to perform various tasks. The best thing of such a setup is that you can roll Windows back to a known good state when it starts to misbehave. I used to have various PCs as well but that drove the electricity bill up considerably.

[Simon]

Well my new laptop has replaced the desktop as well, it can drive my 4K monitor so the desktop is redundant, now I get to only worry about one machine and knoww that what is on it comes with me everwhere I go and the power adapter is 45W, my old processor could use twice that never mind the whole desktop.

[Twoflower]

[...]
I discovered that the power rating at 2.5 GHz is 15 W
[...]

You're not quiet right at the 15W. This processor can be configured up to 25W from the OEM. So it's still within spec (see ark.intel.com).

By the way, especially BOINC could run much faster on later processors as they provide more up to date instruction set compared to the good old 2xxx series (running here a 2600k). The AVX2 could be very useful for scientific workloads. Even if the pure frequency and general IPC is similar.

[shteii01]

On the desktop, i3 were single and dual cores, i5 were quad core, i7 were quad and more cores.  Like you, I am not really into i7, I would have to pull up wiki to find out what exactly difference is between i5 and i7.

I don't know what the difference between the i3, i5 and i7 are, but I suspect the power profile is the same on all as it is common across processors, more speed = more heat, so for more speed more voltage is required.

[shteii01]

I will just turn it off or put it to sleep.  I don't plan to run them 24/7.

I have a general purpose pc, and after a while it gets really cluttered.  The fact that I use some programs rarely and actually forget about them does not help.  Then all the patches and driver updates, etc., and 4 or 5 years later I would reinstall os anyway.

Yeah, the whole i series of intel cpu is like runaway train.  They left me behind when they moved past i5 2xxx series.
Personally I am happy to see that used i5 2xxx series machines (complete machines, just add hdd/sdd) are hitting 100-150 dollar mark.  I might buy a couple of the little ones and set them up for separate tasks, like one machine for cad, one for programming uC, one to play games.

That wastes a lot of energy! Better get one beefy PC and use virtual machines to perform various tasks. The best thing of such a setup is that you can roll Windows back to a known good state when it starts to misbehave. I used to have various PCs as well but that drove the electricity bill up considerably.

[Simon]

[...]
I discovered that the power rating at 2.5 GHz is 15 W
[...]

You're not quiet right at the 15W. This processor can be configured up to 25W from the OEM. So it's still within spec (see ark.intel.com).

By the way, especially BOINC could run much faster on later processors as they provide more up to date instruction set compared to the good old 2xxx series (running here a 2600k). The AVX2 could be very useful for scientific workloads. Even if the pure frequency and general IPC is similar.

Yes it was certainly designed for up to 25 W but a lot of specifications state that it is 2.5 GHz and 15 W. What you then find is you have a processor that gives you a pleasant surprise and can run at up to 3.1 GHz non-stop. But what you are not made aware of usually is the huge power consumption increase. I guess that it works the other way as well and that reducing the core speed further get exponential reductions in power consumption. It's just that this thing is marketed as 2.5 GHz and 15 W. There are some specifications around but do say that it can go up to 25 W. My main point is that for a slight reduction in performance particularly on something you're not too bothered about you can gain a huge saving in power which if anything is an advantage for things like noise that is important to you. Unfortunately the smaller you make a fan but noisier it becomes to move the same amount of air. Fan laws are actually terrible things by the looks of it as bad as processor power consumption laws, so by reducing the heat output you reduce your fan speed. If your computer fan only needs to run at half the speed because the processor is using less power your computer fan will use 1/8 of the power.

[shteii01]

Bigger heatsink is the answer.  Water next.

[...]
I discovered that the power rating at 2.5 GHz is 15 W
[...]

You're not quiet right at the 15W. This processor can be configured up to 25W from the OEM. So it's still within spec (see ark.intel.com).

By the way, especially BOINC could run much faster on later processors as they provide more up to date instruction set compared to the good old 2xxx series (running here a 2600k). The AVX2 could be very useful for scientific workloads. Even if the pure frequency and general IPC is similar.

Yes it was certainly designed for up to 25 W but a lot of specifications state that it is 2.5 GHz and 15 W. What you then find is you have a processor that gives you a pleasant surprise and can run at up to 3.1 GHz non-stop. But what you are not made aware of usually is the huge power consumption increase. I guess that it works the other way as well and that reducing the core speed further get exponential reductions in power consumption. It's just that this thing is marketed as 2.5 GHz and 15 W. There are some specifications around but do say that it can go up to 25 W. My main point is that for a slight reduction in performance particularly on something you're not too bothered about you can gain a huge saving in power which if anything is an advantage for things like noise that is important to you. Unfortunately the smaller you make a fan but noisier it becomes to move the same amount of air. Fan laws are actually terrible things by the looks of it as bad as processor power consumption laws, so by reducing the heat output you reduce your fan speed. If your computer fan only needs to run at half the speed because the processor is using less power your computer fan will use 1/8 of the power.

[Twoflower]

The whole turbo states and power reduction states are a mess to understand. I agree. A lot of performance is dependent on the cooling system. Running long time loads is usually not the target for the thin laptops. I'm not saying that's not possible. But you might suffer from the noise.

The magic words are run-to-idle. The target is to finish a job as fast as possible to reach faster the low power states. Which is for most users the most power efficient approach. With a task that does not end it just runs.

But from my understanding even with the 15W limit the CPU might reach the 3.1GHz. But only one core is used and(!) the GPU on the same processor die is not doing a lot (e.g. not running any 3D applications) and if the temperature is not exceeding certain limits. And in future it will also include the absence of AVX instructions as they drive the power consumption through the roof. The latest Xeon and Xeon-Phi processors already reducing the speed in case of the usage of the vector units (AVX instructions) to stay within the limits.

I used to run several instances of BOINC or Seti@home before that. But it was in times a idling CPU used a lot of power just doing nothing. But since a couple of years the power management reduced the idle power to some 100mW. I read a report that modern laptops draw on idle windows screen @100cd about 3W (including the screen!). That made me reconsider not to run such tasks in background any more.

Water cooling is not very pleasant in a laptop. It does exist for heavy gaming machines. And bigger cooling: I think the manufacturer on Simons laptop is as big as it could fit.

[Simon]

My laptop is HP. It is quite noisy on full load at top speed. But as I said before just reducing down to 2.5 GHz drastically reduces the noise even though I am now running 100%. Yes I am pretty sure that running programs like BOINC does use a lot more powe rbut I run it because I want to help the projects not because I think the power pwill be used anyway.

[nctnico]

The whole turbo states and power reduction states are a mess to understand. I agree. A lot of performance is dependent on the cooling system. Running long time loads is usually not the target for the thin laptops. I'm not saying that's not possible. But you might suffer from the noise.

That is true for sure. I have a Xeon 5-ish something system from Dell and it is totally quiet even at full load. There isn't even a fan on the CPU itself!

[Skimask]

I would have to pull up wiki to find out what exactly difference is between i5 and i7.

Hyperthreading.
i5 = none
i7 = yes

As far as that business about Core 0 being physicall and the rest being virtual...
Ya...No...
Multi core CPUs are just that...CPUs that have multiple physical cores.  Case closed.
The only virtual about any of the cores are those that might hyperthreaded.
Lots of misinformation in thread...lots of it.

[Howardlong]

I would have to pull up wiki to find out what exactly difference is between i5 and i7.

Hyperthreading.
i5 = none
i7 = yes

As far as that business about Core 0 being physicall and the rest being virtual...
Ya...No...
Multi core CPUs are just that...CPUs that have multiple physical cores.  Case closed.
The only virtual about any of the cores are those that might hyperthreaded.
Lots of misinformation in thread...lots of it.

Not quite the case for mobile/laptop processors, where it's all over the place. It's a marketing idiot's wet dream.

[Skimask]

Not quite the case for mobile/laptop processors, where it's all over the place. It's a marketing idiot's wet dream.

Aye...can't argue there.
Clowns just couldn't stick with one formula, one naming convention.  They had to go all apeshit with it.  I'm sure some dickhead got a big pay raise for it.

[hans]

Indeed, Intel's i5/i7 structure has become a mess.
H and desktop parts i5 and i7 are quad core, with i7 having HT.
U parts i5 and i7 are dual core with both HT.

The i7 U is clocked like 100MHz higher than the i5 U part, which manufacturers ask a hefty 100-200$ upgrade price for. But it performs almost identically.
Even more upsetting thing is that this is still a dual core CPU. It's surely is a quick machine for most tasks, until you start 2-3 VM's and quickly run out of CPU horse power.

TDP of 15W is by the way a thermal design power, not actual power consumption. Some of the m-series CPU's are only 3-5W or so, but that is maximum by Intel and some can even go lower. E.g. look at the normal Macbook; it has no fan and uses the chassis for cooling. That machine at full load it will start out using turbo boost but will eventually throttle, unless you watercool the laptop (search Linustechtips youtube channel for that video )

By the way, this turbo nonsense on laptop CPU's can sometimes be a bit obnoxious indeed.
When I use my Lenovo P50 on battery, I often come across this threshold where the CPU can be passively cooled when I'm idle reading a web page article or something. But as soon as I start scrolling in Firefox (which is not the most lightweight browser to be fair), load goes up, CPU fan turns on with a very noticable kick in rpm at first, idles for 5 seconds and then shuts off again.
Maybe I should also seek for a maximum CPU usage on battery dial

[IanB]

Hyperthreading.
i5 = none
i7 = yes

Not quite exactly true. Here is a list of all HT processors:

http://ark.intel.com/search/advanced/?s=t&HyperThreading=true

As far as that business about Core 0 being physicall and the rest being virtual...
Ya...No...
Multi core CPUs are just that...CPUs that have multiple physical cores.  Case closed.
The only virtual about any of the cores are those that might hyperthreaded.
Lots of misinformation in thread...lots of it.

It is rather confusing though, when each physical core shows up as two cores in the task manager. If I have a machine with four physical cores, the system info shows "eight" cores each with its own CPU graph.

[Monkeh]

Of course the thing to remember with these Intel processors is that the virtual core is not actually a processing core it is just extra instruction and data queueing circuitry that while it will slightly increase the performance of the processor at the end of the day once you have hit 50% usage you are in effect at 100%.

Are you at '100%' CPU usage if the instructions you're executing are only using part of the core? No! These are insanely complex cores with a lot of duplication in them. Why wait for the entire core to be free to perform float operations if the current thread isn't doing floats?

Perhaps given the extra efficiencies the virtual cores bring about you could maybe stretch it to 60%.

In real-world usage hyperthreading shows performance boosts much higher than 10%.

Except for those who are convinced it can't be a good thing.

[shteii01]

It is repeatedly claimed to be a 15 W processor. Obviously the extra 600 MHz are very welcome but the power penalty for me does not justify it unless for occasional spurts. Say if I was auto routing a circuit board I would be happy for it to run to that power for a few minutes in order to get the routing done more quickly but not continuously.

Of course the thing to remember with these Intel processors is that the virtual core is not actually a processing core it is just extra instruction and data queueing circuitry that while it will slightly increase the performance of the processor at the end of the day once you have hit 50% usage you are in effect at 100%. Perhaps given the extra efficiencies the virtual cores bring about you could maybe stretch it to 60%. I do in fact have BOINC set to use only two cores for this reason, using more than two cores would just cause a conflict and on my work computer that has an i7 this actually caused absolute mayhem and tasks were failing to complete.

Are you actually saying that you need to monitor and control the CPU cores your work PC uses because you think it wasn't able to pipeline and process instructions without absolute mayhem ensueing and tasks failing to complete?

It sounds like obsessive crazy thinking.

There are billions of processors in the world that just run everything a user throws at them. No fiddling required. My would yours be unable to manage the cores without intervention?

You, obviously, never dealt with BOINC and SETI@Home or similar folks.  I know because I am one of them.

[Simon]

Are you actually saying that you need to monitor and control the CPU cores your work PC uses because you think it wasn't able to pipeline and process instructions without absolute mayhem ensueing and tasks failing to complete?

It sounds like obsessive crazy thinking.

There are billions of processors in the world that just run everything a user throws at them. No fiddling required. My would yours be unable to manage the cores without intervention?

This relates specifically to BOINC which does not seem to have the ability to distinguish between real and virtual processor cores. The result on and I processor is that it tries to carry out twice the amount of tasks that the processor can actually handle because they are intensive processing tasks and hyper- threading does not help. Trying to give a task to a physical core and a task to the virtual core of that physical core is not going to get you to tasks carried out in the world of BOINC. And yes when you have eight tasks trying to use for cause it did get rather problematic. The machine was not always responsive depending on what tasks were working and sometimes crashed. I don't think BOINC is the best written program because it continually tries to switch between tasks to look like it is doing more work than it really is. The result is that not only do you start twice as many tasks as there are processor cores you also then switch to hold another set of twice the amount of tasks as there are processor cores and before you know it there is an over usage of RAM and BOINC can't figure out why tasks are not completing on time. Of course this was some time ago so the updates may have improved things but personally I am not a fan of BOINC itself as a program as it lacks greatly in flexibility.

Of course when you then try to run a 3D CAD program that carries out every bit of graphical visualisation on the processor and RAM rather than the graphics processor although I don't know if the graphics processor could do the job but it is done by the processor in any case this creates absolute mayhem and the screen stutters like mad. Over time I have learnt to fine tune the settings and I limit the amount of cores available to BOINC or set the 3D CAD as an exclusive program as the important thing for the 3D CAD program which I do deem to be rubbish is to have absolute priority access to the RAM memory so that it can carry out all the calculations it needs to carry out when you are rotating models on screen. Of course that computer is quite old now and the RAM is only 1.3 GHz although my new laptop is not that smart on 3D CAD anyway and that has DDR 4 RAM. So the desktop may come useful if I do any serious 3D CAD at home as there is a dedicated high-end graphics card on it and the processor is very fast indeed and it seems that the 3D CAD program sheer brute gigahertz are what give it performance because it is a dumb program and does not seem to take advantage of any particular optimisations.

[Simon]

It is repeatedly claimed to be a 15 W processor. Obviously the extra 600 MHz are very welcome but the power penalty for me does not justify it unless for occasional spurts. Say if I was auto routing a circuit board I would be happy for it to run to that power for a few minutes in order to get the routing done more quickly but not continuously.

Of course the thing to remember with these Intel processors is that the virtual core is not actually a processing core it is just extra instruction and data queueing circuitry that while it will slightly increase the performance of the processor at the end of the day once you have hit 50% usage you are in effect at 100%. Perhaps given the extra efficiencies the virtual cores bring about you could maybe stretch it to 60%. I do in fact have BOINC set to use only two cores for this reason, using more than two cores would just cause a conflict and on my work computer that has an i7 this actually caused absolute mayhem and tasks were failing to complete.

Are you actually saying that you need to monitor and control the CPU cores your work PC uses because you think it wasn't able to pipeline and process instructions without absolute mayhem ensueing and tasks failing to complete?

It sounds like obsessive crazy thinking.

There are billions of processors in the world that just run everything a user throws at them. No fiddling required. My would yours be unable to manage the cores without intervention?

You, obviously, never dealt with BOINC and SETI@Home or similar folks.  I know because I am one of them.

If you mean the people that write the program I find them to be extremely obtuse. The lack of functionality and flexibility is mind-boggling. You would expect such brilliant people to have come up with something better. The idea of switching tasks every sixty minutes is absolutely stupid. I had a thundering row with somebody on their forum about this and he just could not see it because he does not understand other people. I had it installed on my dad computer who used to turn his computer on for an hour at a time every now and then so if every time he turned it on for an hour it was going to then start a new task nothing would ever complete but they just could not see it.

[TiN]

I would call that threat title way too inaccurate. All due respect, you could learn a bit of theory prior to making it. Intel CPUs since very first Core series provide very comprehensive power management features and monitoring , so you can tweak and adjust all aspects of your system as you like, if you spend bit of time learning how those features work. Keywords for googling: EIST, Power Limits PL1, PL2, PL3, Turbo mode, HyperThreading, Affinity, Package Power, TDP. AMD CPUs have very similar amount of features doing same thing, but called differently and working in a bit different way. Same with GPUs...

If you want real cores only always, disable HyperThreading. Benchmark yours apps, perhaps you don't get much benefit from HT anyway.
To keep CPU at "rated" nominal frequency - disable Turbo mode. It's not overclocking and still fully warranted specification by Intel.

One can assign specific app to run on specific cores only, you don't need to install any 3rd-party bodgeware, native OS (assuming windows) have tools to do so. In Windows just run task manager, go to Processes, select your app, select "Set Affinity" in menu. Here you go . You can automatically start app with desired affinity by using simple BAT-file with "start" command.