EST SpeedStep with E2140 shows wrong frequencies

[Daniel Duerr]

Thanks for the reply, John.

I did look into the Cx states as well and I'm only offered C1 (which  
I've switched to C1E through the BIOS).  I'm running on a relatively  
new Supermicro server motherboard which, if price/features were any  
indicator, would support the multiple Cx states offered by today's  
CPUs.  So, I'm wondering if the reason why I am only seeing one state  
is a limitation of my CPU or a limitation of the motherboard.  I did  
do some research on Intel's site yesterday and it appears that my CPU  
only supports C1/C1E.

Do you have any thoughts on this and/or any ideas how I might go about  
figuring out where the limiting factor is here?  If I can confirm that  
my motherboard/BIOS supports the full range of Cx states then I'll  
focus on finding a better CPU with more options.

Also, do you have any recommended approaches for generating test CPU  
load (say 90-100%) so that I can compare the system power consumption  
to when it's idle?  If I'm already indeed idling (in C1E state) at the  
8W consumption Intel suggests then I don't need any more Cx states  
because that is low enough for me.

Thanks again for your help!

Cheers,
Daniel


On Feb 26, 2009, at 5:16 AM, John Baldwin wrote:

> On Wednesday 25 February 2009 7:45:35 pm Daniel Duerr wrote:
>> Hello,
>>
>> I have been following the developments (or lack thereof) around EST
>> and the more recent Intel Dual-core CPUs and was very happy to
>> discover some new results with the latest FreeBSD 7.1 versions,
>> specifically Gabriel Lavoie's recent posts about his success with the
>> E5200 CPU.  My CPU is an Intel Pentium Dual 65nm E2140 @ 1.6GHz which
>> is supposed to use under 8W of power when idled down with EST.  Since
>> this server is always on, this power savings would be wonderful.
>>
>> Today, I decided to give it another shot and updated my 7.1-STABLE
>> amd64 system to the current sources.  I re-enabled the cpufreq driver
>> in my /boot/loader.conf and now, upon bootup, no longer get the
>> message about EST not recognizing my CPU which seems like a great  
>> step
>> forward.  Furthermore, EST is actually attaching to the cpufreq
>> subsystem and populating frequency information as it should.
>> Unfortunately I am experiencing two issues though:
>> 1) the resulting frequency information picked up by EST don't seem
>> complete/accurate for my CPU
>> 2) the behavior changes further if I disable the multiple cores in my
>> BIOS
>
> We rely on the BIOS to tell us the available speed settings via  
> ACPI.  Also,
> for idle consumption the bigger gain will come from using higher Cx  
> states
> rather than using throttling.
>
> -- 
> John Baldwin



--
daniel duerr   |   president   |   gizmo creative
dd at gizmocreative.com  |  +1 (831) 621-1710 x103