Re: [dwm] [OT] frequency scaling and power consumption
On Sat, 09 May 2009, Preben Randhol wrote: > On Sat, 9 May 2009 13:56:09 +0800 > bill lam wrote: > > > Thanks all for confirmation. Since I only have a desktop, no > > notebook/netbook ;-( it is somehow difficult to verify the > > improvement. I now lock the cpu frequency to 1.1GHZ (half of the > > original 2.1G). > > Why on earth do you want to do this to a desktop computer? It will have > next to no effect on your power consumption. I mean you whole system > will draw the main part of the power. Anyway, put the power saving so > that the CPU scales with load. This makes more sense. And turn down > the brightness on your monitor, that saves power too. I got a wattage/current electricity meter to measure the actual power consumption. assuming power factor=1, unit in watts BIOS LOCKED AT ... 1.1GHZ ...2.2GHZ loading ~3% 5355 mplayer/avi ~16% 6670 ./configure && make ~50% 6983 full load >90% 7593 It appears the saving is only a few watts for light to normal loading. I use reverse video (white on black), the lcd use about 10 watts. -- regards, GPG key 1024D/4434BAB3 2008-08-24 gpg --keyserver subkeys.pgp.net --recv-keys 4434BAB3 唐詩013 王維 送別 下馬飲君酒 問君何所之 君言不得意 歸臥南山陲 但去莫復聞 白雲無盡時
Re: [dwm] [OT] frequency scaling and power consumption
On Sat, May 9, 2009 at 9:57 AM, Matthias-Christian Ott wrote: > It's more efficient to run at full speed and then let the CPU halt (you > have to have a tickless system though) [1-4]. Decreasing power consumption The other important thing I've gathered from PowerTOP is that if you have any background processes which insist on doing work even if there's nothing to do (eg, too short journal writeback times, software that polls rather than waits). -- cheers, dave tweed__ computer vision reasearcher: [email protected] "while having code so boring anyone can maintain it, use Python." -- attempted insult seen on slashdot
Re: [dwm] [OT] frequency scaling and power consumption
On Sat, May 9, 2009 at 7:29 AM, Kurt H Maier wrote: > From a physics standpoint if you're generating less heat you're > consuming less power. > > # Kurt H Maier > > And you generate less heat when you allow the cpu to take advantage of it's high frequencies at high loads, because you can put it back into a lower power state early. It didn't use to be this way though with older cpus, so you should definitely subscribe to the powertop mailinglist if that is of real concern to you.
Re: [dwm] [OT] frequency scaling and power consumption
On Sat, May 09, 2009 at 12:29:50AM -0500, Kurt H Maier wrote: > From a physics standpoint if you're generating less heat you're > consuming less power. That's true, but I don't believe that the reduction of power consumption is proportional to the performance losings, so the performance per watt ratio could be worse. Regards, Matthias-Christian
Re: [dwm] [OT] frequency scaling and power consumption
On Sat, May 09, 2009 at 01:56:09PM +0800, bill lam wrote: > > Personally I have noticed that locking my laptops scaling CPU to the > > lowest frequency does give quite a noticeable improvement to the battery > > life, around an extra hour on top of the usual 4~ hours and reduces the > > temperature enough to make the fan shut off . Just enabling on-demand > > scaling didn't help much as it would scale up to full frequency far to > > often. Even with the CPU locked in lo frequency mode it almost never lags. > > > > > > Thanks all for confirmation. Since I only have a desktop, no > notebook/netbook ;-( it is somehow difficult to verify the > improvement. I now lock the cpu frequency to 1.1GHZ (half of the > original 2.1G). > > Less heat means less power consumption, I guess someone suggested, > (I'm not sure that's why I asked for advise), that it takes more time > to complete the job at lower frequency so that actual power > consumption will in some case increase. It's more efficient to run at full speed and then let the CPU halt (you have to have a tickless system though) [1-4]. Decreasing power consumption of sychronous processors is a really hard problem [5-7]. But anyhow I'm not a physicist or electrical engineer. Regards, Matthias-Christian [1] http://www.codon.org.uk/~mjg59/power/good_practices.html [2] https://www.redhat.com/docs/wp/performancetuning/Power_Management_Guide.pdf [3] http://www.ncsu.edu/wcae/ISCA2007/p52-suarez.pdf [4] http://lists.us.dell.com/pipermail/linux-poweredge/2007-December/033900.html [5] http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3276&p=6 [6] http://patmos2001.eivd.ch/program/Repro%5CArt_10_1.pdf [7] http://async.org.uk/ukasyncforum14/forum14-papers/forum14-moore.pdf [8] http://www.fulcrummicro.com/press_archives/edn_03-0501.pdf
Re: [dwm] [OT] frequency scaling and power consumption
On Sat, 9 May 2009 13:56:09 +0800 bill lam wrote: > Thanks all for confirmation. Since I only have a desktop, no > notebook/netbook ;-( it is somehow difficult to verify the > improvement. I now lock the cpu frequency to 1.1GHZ (half of the > original 2.1G). Why on earth do you want to do this to a desktop computer? It will have next to no effect on your power consumption. I mean you whole system will draw the main part of the power. Anyway, put the power saving so that the CPU scales with load. This makes more sense. And turn down the brightness on your monitor, that saves power too. HTH -- Preben Randhol http://wee-free-lore.blogspot.com/
Re: [dwm] [OT] frequency scaling and power consumption
> Personally I have noticed that locking my laptops scaling CPU to the > lowest frequency does give quite a noticeable improvement to the battery > life, around an extra hour on top of the usual 4~ hours and reduces the > temperature enough to make the fan shut off . Just enabling on-demand > scaling didn't help much as it would scale up to full frequency far to > often. Even with the CPU locked in lo frequency mode it almost never lags. > > Thanks all for confirmation. Since I only have a desktop, no notebook/netbook ;-( it is somehow difficult to verify the improvement. I now lock the cpu frequency to 1.1GHZ (half of the original 2.1G). Less heat means less power consumption, I guess someone suggested, (I'm not sure that's why I asked for advise), that it takes more time to complete the job at lower frequency so that actual power consumption will in some case increase. -- regards, GPG key 1024D/4434BAB3 2008-08-24 gpg --keyserver subkeys.pgp.net --recv-keys 4434BAB3 唐詩187 杜甫 登樓 花近高樓傷客心 萬方多難此登臨 錦江春色來天地 玉壘浮雲變古今 北極朝庭終不改 西山寇盜莫相侵 可憐後主還祠廟 日暮聊為梁父吟
Re: [dwm] [OT] frequency scaling and power consumption
>From a physics standpoint if you're generating less heat you're consuming less power. # Kurt H Maier
Re: [dwm] [OT] frequency scaling and power consumption
On Fri, 8 May 2009 16:08:12 +0800 bill lam wrote: > BIOS support choosing a smaller multipliers to reduce cpu frequency. > linux also supports frequency scaling such powernowd. Some google > page said cpu throttling can not reduce power consumption. My > experience is that it seems to lower temperature. If it can also > reduce power consumption, I'm willing to save money by running cpu at > half of its current frequency. Any idea. > same experience. But I let the software decrease the cpu throttling stepwise on load. I mean not jumping straight between 800MHz and 1.6GHz nor force it to be 800MHz. For powersaving I gained 1 hour (from 5 hours to 6 hours) by blacklisting all kernel modules I don't want to be started on boot on my Asus 1000H. E.g. not staring web camera support on startup. Experiment with this on your computer. Best wishes Preben
Re: [dwm] [OT] frequency scaling and power consumption
> On Fri, May 8, 2009 at 9:08 AM, bill lam wrote: >> BIOS support choosing a smaller multipliers to reduce cpu frequency. >> linux also supports frequency scaling such powernowd. Some google >> page said cpu throttling can not reduce power consumption. My >> experience is that it seems to lower temperature. If it can also >> reduce power consumption, I'm willing to save money by running cpu at >> half of its current frequency. Any idea. > > My understanding is that power usage scales nonlinearly with CPU > frequency, and in particular having twice the frequency doesn't > require quite as much as double the power. So IF your OS can put your > PC into proper "sleep" states whent here's nothing to do (and that's > the big IF), the PC will use less energy in total by running at full > speed when you have work to do and then going into a sleep state > rather than taking twice as long to do the work at half the frequency. > So I'd expect you'd probably get more "energy usage reduction" from > getting rid of any services/device drivers/etc that stop the PC going > to sleep than from manually reducing the frequency. (If you're using > Linux, PowerTOP ( > > www.lesswatts.org/projects/powertop/ > > ) is an attempt to provide a way to at least see what's causing > wake-ups, even if it doesn't necessarily show how to solve them.) >> -- Personally I have noticed that locking my laptops scaling CPU to the lowest frequency does give quite a noticeable improvement to the battery life, around an extra hour on top of the usual 4~ hours and reduces the temperature enough to make the fan shut off . Just enabling on-demand scaling didn't help much as it would scale up to full frequency far to often. Even with the CPU locked in lo frequency mode it almost never lags.
Re: [dwm] [OT] frequency scaling and power consumption
On Fri, May 8, 2009 at 9:08 AM, bill lam wrote: > BIOS support choosing a smaller multipliers to reduce cpu frequency. > linux also supports frequency scaling such powernowd. Some google > page said cpu throttling can not reduce power consumption. My > experience is that it seems to lower temperature. If it can also > reduce power consumption, I'm willing to save money by running cpu at > half of its current frequency. Any idea. My understanding is that power usage scales nonlinearly with CPU frequency, and in particular having twice the frequency doesn't require quite as much as double the power. So IF your OS can put your PC into proper "sleep" states whent here's nothing to do (and that's the big IF), the PC will use less energy in total by running at full speed when you have work to do and then going into a sleep state rather than taking twice as long to do the work at half the frequency. So I'd expect you'd probably get more "energy usage reduction" from getting rid of any services/device drivers/etc that stop the PC going to sleep than from manually reducing the frequency. (If you're using Linux, PowerTOP ( www.lesswatts.org/projects/powertop/ ) is an attempt to provide a way to at least see what's causing wake-ups, even if it doesn't necessarily show how to solve them.) > -- > regards, > > GPG key 1024D/4434BAB3 2008-08-24 > gpg --keyserver subkeys.pgp.net --recv-keys 4434BAB3 > 唐詩094 杜審言 和晉陵路丞早春遊望 >獨有宦遊人 偏驚物候新 雲霞出海曙 梅柳渡江春 >淑氣催黃鳥 晴光轉綠蘋 忽聞歌古調 歸思欲霑巾 > > -- cheers, dave tweed__ computer vision reasearcher: [email protected] "while having code so boring anyone can maintain it, use Python." -- attempted insult seen on slashdot
