On Thu, Jan 09, 2014 at 03:47:22PM +0100, Olliver Schinagl wrote: >I think its pretty fair to conclude, that power draw when IDLE is >almost the same on min freq and max freq. During load it does matter >but why would you run your CPU lower during load? So for the default >governor, performance without a doubt and if you insist on ondemand, >408 for the minimum is quite alright. > >Interesting result to say the least.
a few years ago I measured the same thing on intel server chips (nehalem and older xeons in HPC clusters). server idle power @ 200Mhz == server idle power @ 3GHz. likely internal blocks in the chip are smart and clock themselves down/off when they detect they're idle. on the intel chips (and it looks like on allwinner too) this happens regardless of any OS cpufreq settings. in reality (which isn't a case commonly covered by benchmarks :-) the main effect of lower MHz cpu governors is probably to make apps that are spinwaiting for i/o or network use less power. unfortunately lower MHz can also mean higher latency from the i/o and network subsystems, so it may not be a win. my conclusion (in the HPC workload case) was that max MHz was clearly the best choice. IMHO it's hard to construct cases where max MHz is wrong... (+/- chip might melt) BTW, intel have gradually been improving the whole-server idle/max power usage ratio - largely by moving functionality off the board and into the cpu die - SoC style. the ratio used to be about 1/2 on Core2 era servers, was about 1/3 on Nehalem, and I'd guess it is now probably closer to 1/4. from your numbers it looks like allwinner/cubie2 is 0.34/.9 ~= 0.38 which isn't too bad, but isn't great either. cheers, robin -- You received this message because you are subscribed to the Google Groups "linux-sunxi" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/groups/opt_out.
