Gok,strange problem here.I tried to make a nandroid backup(a normal one,not of the secondary rom) and after backing up data it gave me an error message saying that there was an error when backing up .secondrom/data or something.Dafuq?
EDIT:What chance is there that the kernel panics I was and am still getting(although not nearly as often now) are related to overclocking while having changed the internal voltages?With a little higher that stock internal voltages,my phone is much stabler than it was with low voltages at high frequencies,but if kernel panics are related to that in any way,I will return to my previous quasi-unstable setup.
As said, ondemand based governors do not visit steps so in an ideal situation there will be no increased transitions at all; if the governor decides to stay at the same frequency. But I doubt this is a real-world situation, the governor would jump either way or another to another frequency unless under heavy load. You have a transition time of 100µS compared to a default sampling rate of 100000µS or minimum 20000µS so we are talking about a loss of computing time two orders of magnitude smaller than a sampling period, versus a possible power gain in clocking of in the most optimistic case of 50% and worst case 7%.
I did not say they jump to the highest frequency too much, they jump as much as they need to. You have smoothing of load spikes through larger sampling periods and as you said through smooth scaling. I think interactive governors are a bad choice for this device and I haven't seen the point in using them; they may be good for battery use but the lags they introduce even with a tight frequency table is too noticeable for my taste. Use a smaller frequency table for all you want and then use them and disregard the whole discussion. I will be sticking with ondemand though. Frequency steps in the GS3 have no correlation with the number of cores, so I don't know why you'd bring that up, it doesn't change anything in terms of scaling, as in that regard Pegasus has the same logic as classic ondemand.
100mV is almost guaranteed for 100 and 200MHz, the rest depends on the device. I wrote it just as an example of measurement, it's up to the user to decide what to do with it. I also measured underclocking and undervolting GPU, it has almost not measurable difference in light load. The Mali GPU also has internal power management that I think clock-gates part of the digital core when under low load, you can see this in the drivers. Undervolting GPU under load in games will bring a big decrease in power though, undervolting 200mV will bring down the total power consumption down 15% (that includes the whole device, screen on, CPU locked at 800), so the percentage is even greater I one would be able to solely measure the GPU.
I don't see what you want to say. The W/GHz value is only a theoretical calculation derived from his measurements under full load which is not realistic in a low load scenario. For playing music our devices don't even hit deep idle, ever. At least mine didn't, no matter what frequency you'd be playing music at, 100, 200 or 500MHz, the load is too fine-grained to let the CPU enter any clock-gating. There is no case in which 100MHz stays at high load because the governor will simply up the frequency and that situation will never realistically happen.
|battery life, cwm touch 18.104.22.168, dualboot, extweaks with sound booster, ics, jelly bean, kernel that walk in the path of heaven, legendary, stable, stweaks|
|Thread Tools||Search this Thread|