Forgot to mention SPEC Power: http://www.spec.org/power_ssj2008/
On Mon, Oct 11, 2010 at 9:53 AM, Stephen Dawson-Haggerty <[email protected]> wrote: > Also jumping on this thread late. Does anyone know of a database > collecting this information? > > We have a small databased locally that we might be willing to share -- > have to ask around. Agree it's complicated, but usually the basics > get you pretty far. We try Linux with (a) "idle" and (b) serving > lighthttpd requests to saturation. We're interested in whole-platform > consumption, although we've broken it down to components in the > past... > > Steve > > On Mon, Oct 11, 2010 at 9:45 AM, Auke Kok <[email protected]> wrote: >> On 10/8/2010 12:58 AM, Paul van der Vlis wrote: >>> >>> Auke Kok wrote: >>>> >>>> On 10/03/10 10:44, Paul van der Vlis wrote: >>>>> >>>>> Hello, >>>>> >>>>> I am looking for a list with the idle power consumption of CPU's. >>>>> >>>>> There are many lists with the TDP, but that's something different. I >>>>> have no problems with power-consumption when the machine is really busy, >>>>> but I hate it when there is a big consumption when there is nothing to >>>>> do. >>>> >>>> This is a bit harder to generate than TDP etc. Do you want the power >>>> consumption of your CPU *package*? Or just one core? Or even a thread? >>>> Or do you want wall-equivalent power? What's Idle? Accounting for C4/5/6 >>>> deep C-states? >>>> >>>> etc... >>> >>> (I did not get this message by mail, but I saw it in the archive.) >>> >>> What I do now is to measure the powerconsumption of a CPU inclusive >>> mainbord, power-supply, memory and disk. >>> >>> I realise that it would be better if I would have the consumption of the >>> CPU package alone. But I would need also information then about the >>> mainboard. >>> >>> With "idle" I mean a state without workload, but where everything is >>> functional. >>> >>> I understand it is not easy to find a good way, but I think you also >>> understand it is important to have this kind of information. >>> >>> Maybe it would be good to measure a "reaction time" in the idle state >>> too. E.g. of a webserver. >> >> that information is actually relatively easy to retrieve since every C-state >> has a specific "exit latency" which is known (the scheduler uses this to >> decide which C-state to enter for instance). So, C-state statistics can be >> used to calculate average return-from-idle latency trivially... >> >> Auke >> _______________________________________________ >> Discuss mailing list >> [email protected] >> http://lists.lesswatts.org/listinfo/discuss >> > > > > -- > stephen dawson-haggerty > http://cs.berkeley.edu/~stevedh > uc berkeley wireless and embedded systems lab > berkeley, ca 94720 > -- stephen dawson-haggerty http://cs.berkeley.edu/~stevedh uc berkeley wireless and embedded systems lab berkeley, ca 94720 _______________________________________________ Discuss mailing list [email protected] http://lists.lesswatts.org/listinfo/discuss
