cpufreq: CPU is faster with ondemand governor than with userspace

Gusar · 2012-11-04 14:14:25

ernetas wrote:

The Wiki is the proof.

The wiki just states that it is so. "Proof" would be links to measurements being made with various settings. I'm not disputing that tweaking the governor has an effect. It's just that you were berating me about proof, and yet you make claims while not providing any.

ernetas wrote:

By saying "CPUs power management was not intended to be under cpufreq" I mean, that Intel did not gave support for the creation of this utility.

You just repeated the same nonsensical sentence.

ernetas wrote:

Yes. Did I say it in any significantly different way that you had to repeat it?
Let me quote your words again:
Gusar wrote:
ernetas wrote:
It is exactly the same as userspace governor - ondemand governor ALSO fixes a frequency, but it "refixes" it when needed.
That makes no sense.

I stand by what I said - it makes no sense. If ondemand is constantly "refixing", then by logic things are not fixed, they are fluid.

ernetas wrote:

It's not funny, but a waste of time. Now you say that I DID NOT lock the frequency at 2.301, even though before you said it vice versa:
Gusar wrote:
ernetas wrote:
Are you trying to tell me that without CPU frequency scaling in kernel, you can't have Intel Turbo Boost working?
Why does this surprise you?

Err, my statement is still true - without freq scaling, you cannot have turbo working. You still need to have the acpi_cpufreq module loaded, and things need to specifically be set so that turbo can work. I just didn't know the userspace governor has such a setting. But if you set userspace to a fixed freq, then turbo indeed does not work.

ernetas wrote:

So, according to you, ArchLinux Wiki is bullshit.

It is, actually. The sentence "For Laptops or other mobile systems, the conservative governor can possibly provide significant savings in power consumption" is completely wrong. Read the link I already gave: http://mjg59.livejournal.com/88608.html. That sentence I got just from a quick glance, who knows what else the wiki has wrong. I don't have a wiki account yet, though it seems one of these days I'll have to create one.

ernetas wrote:

Hypothetical? Any bigger torrent tracker has MySQL servers running on load higher that the number of cores. Any bigger cheap hosting service has servers running on load higher than the number of cores.

These CPUs run at 100% *all* the time? I'd like to see that - do you have reports, like maybe blog posts with graphs and such? I'm not denying it isn't true. Though I would've thought serving is more of an I/O thing. So, you know, show me proof. See, instead of spinning words around like crazy, I'm fully willing to learn something new.

ernetas wrote:

At last, I can just write a scientific simulation program in C++ and make my CPU loaded at 100% for all time. BOINC also does that.

How is writing a simulation program not creating a scenario just so that you can make a point? Of course there are long-running full load tasks. But you won't be running them all the time on the machine that is the main topic of this post - your laptop. You won't even be running them on many servers.

ernetas wrote:

Overall:
Before this argument, you said that I cannot have Turbo on with userspace governor and that's why ffmpeg is performing worse while using this governor. But now you know that I can use userspace and Turbo Boost. End of argument. I don't have time to read through this bullshit of word play between us.

It's so nice that you call me trying to educate you on power management "bullshit". I didn't know userspace has a mode where turbo is allowed. I learned something. Did you learn anything? Like why FFmpeg is slower with userspace, despite using turbo? Did you learn that you should not be using userspace?

Last edited by Gusar (2012-11-04 14:17:02)

CarbonChauvinist · 2012-11-04 15:43:24

one of the funnier pissing contests I've seen recently

ewaller · 2012-11-04 18:43:51

CarbonChauvinist wrote:

one of the funnier pissing contests I've seen recently

Now that made me laugh.

Putting aside marketing terms and other nonsense, here are a few truths:

The power dissipated by a processor is a strong function of the clock speed. In truth, they only draw current at a clock edge.
If one does not take the heat out as fast as you put it in, the processor temperature will increase.
To get the heat out faster, you have to provide additional cooling.
Additional cooling can come from the increased temperature differential between the junction temperatures and ambient air.
Generally, you provide additional cooling by providing increased ambient airflow.
Cooling a computer does place a heat load on the room.
It is possible to clock a processor beyond the frequency where it can be cooled for very short durations due to the thermal mass of the processor. Insert marketing term here.
If the processor is not doing anything useful, it spends most of its time either sleeping, or in the idle loop. During these times, the processor clock can be slowed way down, or even paused.

A processor running at a constant clock frequency and a constant core voltage will draw power at a constant rate.

How all of this ties into the tools (and this thread) is left as an exercise to the reader. I just wanted to dispel any myths.

cfr · 2012-11-08 03:59:35

@ewaller,
May I ask: what is a clock edge? (I can see that Big Ben would have edges but that thought is remarkably unenlightening...)

WonderWoofy · 2012-11-08 04:02:06

I think it is a reference to the top and bottom of the sinusoidal wave.

2ManyDogs · 2012-11-08 04:17:52

https://en.wikipedia.org/wiki/Clock_edge

Wikipedia wrote:

In electronics, a signal edge is a transition in a digital signal either from low to high (0 to 1) or from high to low (1 to 0). It is called an "edge" because the square wave which represents a signal has edges at those points.

https://en.wikipedia.org/wiki/CPU_core_voltage

Wikipedia wrote:

The CPU core voltage (VCORE) is the power supply voltage supplied to the CPU (which is a digital circuit), GPU, or other device containing a processing core. The amount of power a CPU uses, and thus the amount of heat it dissipates, is the product of this voltage and the current it draws. In modern CPUs, which are made using CMOS, the current is almost proportional to the clock speed, the CPU drawing almost no current between clock cycles.

Last edited by 2ManyDogs (2012-11-08 04:22:35)

cfr · 2012-11-08 16:35:06

Thanks. Combining the two answers makes sense to me.

Arch Linux

#26 2012-11-04 14:14:25

Re: cpufreq: CPU is faster with ondemand governor than with userspace

#27 2012-11-04 15:43:24

Re: cpufreq: CPU is faster with ondemand governor than with userspace

#28 2012-11-04 18:43:51

Re: cpufreq: CPU is faster with ondemand governor than with userspace

#29 2012-11-08 03:59:35

Re: cpufreq: CPU is faster with ondemand governor than with userspace

#30 2012-11-08 04:02:06

Re: cpufreq: CPU is faster with ondemand governor than with userspace

#31 2012-11-08 04:17:52

Re: cpufreq: CPU is faster with ondemand governor than with userspace

#32 2012-11-08 16:35:06

Re: cpufreq: CPU is faster with ondemand governor than with userspace

Board footer