re: acpicpu slowing down computer [was Re: firefox slowness -- X server problem?]

[matthew green <mrg%eterna.com.au@localhost>]

> On Thu, Aug 19, 2010 at 11:04:30PM +0200, Jens Rehsack wrote:
> > But how do you explain the test result? My firefox is really
> > slow - it takes several seconds switching between windows.
> > And if I understood Thomas correctly, on his system it's the
> > same. After removing acpicpu(4) it's quick as it was before.
> > Side effect?
> 
> If you want to help in testing, 'boot -x' with acpicpu(4) enabled but
> options(4) ENHANCED_SPEEDSTEP disabled. And use the latest sources. Report
> results to me preferably in private mail.
> 
> What CPU is this? One reason for this may be that on new Intel Nehalem CPUs
> the highest frequency (P0-state) enables the so-called Turbo Boost. But the
> P1-state is the "real" highest frequency.
> 
> You could see something like:
> 
>       acpicpu0: P0: FFH, lat  10 us, pow 35000 mW, 2534 MHz
>       acpicpu0: P1: FFH, lat  10 us, pow 35000 mW, 2533 MHz
> 
> There is a 1 MHz difference between P0 and P1 with Turbo Boost, which, if
> enabled, may either reduce or increase the frequencies behind our back based
> on thermal conditions. It overclocks. (This could happen also with the
> existing ENHANCED_SPEEDSTEP, but it would require a fair amount of good luck.)

it won't happen very much with netbsd today because there's usually only
a very limited boost when all cores are in C1 or higher (and we don't do
anything lower yet.)  eg, my i7 will only boost upto 8% when 3 or 4 (all)
cores are active, 50% when 2 cores are active, and 75% if only 1 core is
active.  however, even if this were happening, it wouldn't explain why
firefox is taking a lot of cpu and being very slow.  if the cpu was turbo
boosting it would be *better* not worse, right?


.mrg.