G
GuestK00471
Guest
So I've been trying to find an answer to this
A commit in Linux 3.12 changed ondemand gov. Behavior below the up threshold to calculating freq based on load and once it hits up threshold it clocks to max. My thought is , why not just 100% load based?
Is it the potential performance hit as it would take high load to hit max freq?
Many modern cpu are able to switch freq very quickly so is the speed of freq switching or the amount of it an issue?
Is it the idea of race to idle and if you aren't peaking and getting the task done asap that's bad?
I'm trying it out atm on my M8 with a stripped and modified conservative gov. and am just curious if there is any reason I shouldn't do this? I essentially set sample rate and that determines how fast it responds. I've notice a small hit to benchmarks but I don't see any other real disadvantage.
Thoughts? C:
A commit in Linux 3.12 changed ondemand gov. Behavior below the up threshold to calculating freq based on load and once it hits up threshold it clocks to max. My thought is , why not just 100% load based?
Is it the potential performance hit as it would take high load to hit max freq?
Many modern cpu are able to switch freq very quickly so is the speed of freq switching or the amount of it an issue?
Is it the idea of race to idle and if you aren't peaking and getting the task done asap that's bad?
I'm trying it out atm on my M8 with a stripped and modified conservative gov. and am just curious if there is any reason I shouldn't do this? I essentially set sample rate and that determines how fast it responds. I've notice a small hit to benchmarks but I don't see any other real disadvantage.
Thoughts? C:
