EST (Enhanced Intel SpeedStep(R) Technology) on amd64

[Alexander Motin]

Gabriel Lavoie wrote:
> 2009/1/9 Alexander Motin <mav at freebsd.org>:
>> Gabriel Lavoie wrote:
>>> Another question. Any reason why powerd doesn't use
>>> dev.est.0.freq_settings when it is available instead of
>>> dev.cpu.0.freq_levels?
>>>
>>> On my system:
>>> dev.cpu.0.freq_levels: 2500/88000 2187/77000 2000/47608 1750/41657
>>> 1600/44616 1400/39039 1200/41800 1050/36575 900/31350 750/26125
>>> 600/20900 450/15675 300/10450 150/5225
>>> dev.est.0.freq_settings: 2500/88000 2000/47608 1600/44616 1200/41800
>> dev.cpu.0.freq_levels is just a mix of est and p4tcc levels. By default
>> powerd uses all of them. If you really wish, you can disable p4tcc with:
>> hint.p4tcc.0.disabled=1
>> hint.acpi_throttle.0.disabled=1
>> then dev.cpu.0.freq_levels will be equal to dev.est.0.freq_settings.
> 
> Working like a charm! I wasn't sure so I added both 0 and 1 for both
> cores and it seems to be correct. I had to add them in loader.conf or
> device.hints? Put them in loader.conf and it seems to be OK.

Whatever you prefer,

>>> If I don't lower the polling time of powerd to 100ms, my system
>>> becomes way too much unresponsive because powerd takes too much time
>>> to increase the frequency, step by step and there are a lot of
>>> settings with dev.est.0.freq_settings (14). With
>>> dev.est.0.freq_settings, the minimal setting is high enough so the
>>> system stays responsive and powerd would bring it up to max frequency
>>> quickly enough, even if the polling time is still kept at 500ms. This
>>> would work more like Windows or Linux where the lowest frequency at
>>> which the CPU will drop is the lowest EIST gives (here 1200 MHz).
>> I have just merged updated powerd to 7-STABLE to address this issue.
> 
> If I run release, it means it will be available in 7.2?

7.2 and 8.0.

-- 
Alexander Motin