Thread: Overclock the N900?
View Single Post
MONVMENTVM is offline MONVMENTVM
2010-06-30 , 12:14
Posts: 145 | Thanked: 91 times | Joined on Jun 2010
#4561
Originally Posted by arbitrabbit View Post
Even while playing Music or running a movie, the CPU is not ON all the time. A lot of that is done via the DSP anyways, so your CPU would remain idle a significant amount of time (as an example, play an mp3 and run Conky and you would see that the CPU is usually at 500MHz if you are running ideal). So while the test is valid, the premise isn't. Also, to sample current properly, unfortunately once every second isn't enough. You are talking more like 30 times a second, given how many times the CPU is waking up every second.

Having said that, there is definitely a compromise frequency, which would vary from device to device. So for ideal, it could be 600Mhz but for LV, it could be lower.
One sample per second is enough in this test because I feed the CPU with constant load. The variation between the results usually is maximum +/- 10mA... if i raise the time span from 1 minute to several minutes or even an hour, then the variation should be even less and we should achieve even more precise results (but you still get a good idea from these results).

Polling 30 times a second should basically be no problem but I assume that this alone would possibly create too much CPU load and therefore affect the result.

You are right that while playing music, the CPU stays mainly at 500MHz with ideal configuration, which is because CPU load stays usually around 20 - 30 %. But that wasn't necessarily what I was testing for, as I was setting the frequencies to a fixed value to check WHICH frequency is the most efficient in terms of battery consumption. So I don't really get your point. Besides if you check powertop while playing music you'll see that the CPU never gets into its deepest sleep state C3 (which again is slightly offtopic here).