powernowd and friends

[Warren Togami]

Behdad Esfahbod wrote:
> 
> The logical conclusion is that, modulating CPU speed *on demand*
> means nothing.  No matter what speed you *think* your CPU is
> working on, your CPU is smart enough to consume enough power to
> complete the job.
> 
> 
> My conclusion is that, CPU speed modulation on demand is as wrong
> as turning lights off when in blackout, and turn on again when
> you are back.

I know absolutely nothing about Intel processors, but here are my 
observations of my 1GHz Athlon4 processor in my Sony Vaio FXA-36.

Running older kernels without ACPI, temperatures are always extremely 
hot, sometimes reaching 70C.  Back in those days with APM kernels, I 
used the crude but effective lvcool program which runs HLT in otherwise 
idle cycles.  It is a userspace process that is niced to be very low 
priority.  ACPI kernels seem to have this effect by default, with 
seemingly less overhead and no longer appearing as using 100% CPU all 
the time like the old lvcool method.  Using either lvcool or ACPI would 
bring temperatures within the 58-64C range.

The 10 ACPI throttle states seem to be effective in slowing down the 
system, but some crude measurements have showed seemingly ZERO 
DIFFERENCE in temperature or power usage.  Thus I have ignored tweaking 
that interface.

Only recently did cpufreq begin working with K7 processors.  Running at 
500MHz I have regular operating temperatures in the 46-52C range and 
extended battery life.  After playing with cpufreq for a day, it was too 
much of a hassle for me to tweak manually so I pretty much ignored it 
until I discovered powernowd.  I like powernowd because I get the 
benefits of low temperatures and extended battery life, but it 
automatically jumps to full speed if I do processor intensive things 
like compiling.

The vast majority of the time however I am only using my laptop to ssh 
into other boxes, so I like that it automatically drops to minimum speed 
where my regular desktop apps are mostly idle cycles, with the exception 
of brief jumps when large applications are loading.

Anecdotally I do not "feel" the effects of the slower speed, probably 
because with powernowd default settings it has a hair-trigger to jump to 
maximum speed when the CPU when it is needed.

https://bugzilla.fedora.us/show_bug.cgi?id=520
fedora.us powernowd package

Warren