do Pentium4 & PentiumM consume about the same at their lowest possible freq?

do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Roberto De Leo]

Hi all,
I am not sure this is the right place where posing my question, if not 
please suggest me a better one!

I come to the question.
I happily used the speedstep kernel module with my centrino laptop and 
it really made a difference in terms of battery duration (actually now I 
have a celeron-m based laptop and the difference does not seem so 
evident, but I still have little statistic on it).
Now I need to replace my main home 'desktop' PC and I am tempted to buy 
a PentiumM-based minit-itx MB (e.g. the AOpen i945GTt-VFA) in order to 
decrease the power bill (my PC is on 24/7), but since mini-itx MBs are 
quite expensive I'd prefer first to know the following: can the  "Intel 
Pentium 4 clock modulation" kernel module (or maybe the "AMD 
Opteron/Athlon64 PowerNow!" one) bring the corresponding CPUs to consume 
as little as the "Intel Enhanced SpeedStep" does for a centrino? This 
info is quite important but I could not find it anywhere!

Thanks,
 Roberto

Re: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Len Brown]

On Friday 03 August 2007 12:56, Roberto De Leo wrote:
> Hi all,
> I am not sure this is the right place where posing my question, if not 
> please suggest me a better one!
> 
> I come to the question.
> I happily used the speedstep kernel module with my centrino laptop and 
> it really made a difference in terms of battery duration (actually now I 
> have a celeron-m based laptop and the difference does not seem so 
> evident, but I still have little statistic on it).
> Now I need to replace my main home 'desktop' PC and I am tempted to buy 
> a PentiumM-based minit-itx MB (e.g. the AOpen i945GTt-VFA) in order to 
> decrease the power bill (my PC is on 24/7), but since mini-itx MBs are 
> quite expensive I'd prefer first to know the following: can the  "Intel 
> Pentium 4 clock modulation" kernel module (or maybe the "AMD 
> Opteron/Athlon64 PowerNow!" one) bring the corresponding CPUs to consume 
> as little as the "Intel Enhanced SpeedStep" does for a centrino? This 
> info is quite important but I could not find it anywhere!

In general, no -- you are better off with a mobile-optimized processor.

indeed, under many conditions, using pentium 4 clock modulation will
actually increase energy use, rather than decrease it.

-Len

Re: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Roberto De Leo]

> indeed, under many conditions, using pentium 4 clock modulation will
> actually increase energy use, rather than decrease it.
>
Jeez, that's odd!
Then maybe this is the reason why I don't notice any particular 
improvement when I slow down my celeron-m, which uses indeed the P4 
modulation sw rather than the speedstep centrino. But then, why slowing 
down the CPU at all if it doesn't save you any watt? I guess I'll stop 
using it.
Thanks a lot for the preciuos info,
Roberto

RE: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Pallipadi, Venkatesh]

 

>-----Original Message-----
>From: cpufreq-bounces@lists.linux.org.uk 
>[mailto:cpufreq-bounces@lists.linux.org.uk] On Behalf Of Roberto De Leo
>Sent: Friday, August 03, 2007 1:16 PM
>To: cpufreq@lists.linux.org.uk
>Subject: Re: do Pentium4 & PentiumM consume about the same at 
>their lowest possible freq?
>
>
>
>> indeed, under many conditions, using pentium 4 clock modulation will
>> actually increase energy use, rather than decrease it.
>>
>Jeez, that's odd!
>Then maybe this is the reason why I don't notice any particular 
>improvement when I slow down my celeron-m, which uses indeed the P4 
>modulation sw rather than the speedstep centrino. But then, 
>why slowing 
>down the CPU at all if it doesn't save you any watt? I guess I'll stop 
>using it.

It is meant to be used under thermal condition where slowing down CPU
will result in reducing the temperature and not for power savings in
normal conditions.

Thanks,
Venki

Re: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Roberto De Leo]

>>why slowing down the CPU at all if it doesn't save you any watt? I guess I'll stop 
>>using it.
>>    
>>
>It is meant to be used under thermal condition where slowing down CPU
>will result in reducing the temperature and not for power savings in
>normal conditions.
>
I thought that power consumption and heat dissipation in the CPU were 
more or less proportional but evidently my belief was too naif  :-)

I guess that the same thing holds for the Athlon64 so prolly centrino is 
the only CPU were I can get a substantial power saving by decreasing the 
freq when I don't need much of it (which is almost always), is this 
correct?

Thanks,
 Roberto

Re: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Jarod Wilson]

[-- Attachment #1.1: Type: text/plain, Size: 973 bytes --]

Roberto De Leo wrote:
> 
>>> why slowing down the CPU at all if it doesn't save you any watt? I
>>> guess I'll stop using it.
>>>   
>> It is meant to be used under thermal condition where slowing down CPU
>> will result in reducing the temperature and not for power savings in
>> normal conditions.
>>
> I thought that power consumption and heat dissipation in the CPU were
> more or less proportional but evidently my belief was too naif  :-)
> 
> I guess that the same thing holds for the Athlon64 so prolly centrino is
> the only CPU were I can get a substantial power saving by decreasing the
> freq when I don't need much of it (which is almost always), is this
> correct?

No. AMD64 does the right thing too. The P4 only throttles, the AMD64 and
centrino chips actually scale their frequencies down. Throttling and
frequency scaling are different beasts, with freq scaling being far more
worthwhile.

-- 
Jarod Wilson
jwilson@redhat.com



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 251 bytes --]

[-- Attachment #2: Type: text/plain, Size: 147 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@lists.linux.org.uk
http://lists.linux.org.uk/mailman/listinfo/cpufreq

RE: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Langsdorf, Mark]

> >It is meant to be used under thermal condition where slowing down CPU
> >will result in reducing the temperature and not for power savings in
> >normal conditions.
> >
> I thought that power consumption and heat dissipation in the CPU were 
> more or less proportional but evidently my belief was too naif  :-)
> 
> I guess that the same thing holds for the Athlon64

Most processors reduce power consumption faster than they reduce
frequency, because reducing frequency allows the processor to
also reduce voltage.  There is a corner case that if you only
reduce frequency, you do not get much power reduction, but the
standard powernow-k[7,8] driver reduces both voltage and
frequency.

The only numbers I have handy are a bit old, but here's the
original Opteron/Athlon FX at 2.2 GHz:

Frequency (GHz)	Voltage (V)		Power (W)
2.2			1.5			89
2.0			1.4			69
1.8			1.3			50
1.0			1.1			22

A drop of 50% in frequency cuts maximum power
by more than 66%.  Actual measured power may be
lower, since idle blocks consume less power,
and a processor at minimum pstate may also be
going into C1 (Halt) a good deal of the time.

-Mark Langsdorf
Operating System Research Center
AMD

Re: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Erich Boleyn]

FYI, I work at AMD currently, and worked at Intel in the past.  The
model discussed below is an approximation, and some circuit or process
geek can correct me if they care to.  ;-)


Roberto De Leo <deleo@unica.it> wrote:

> I thought that power consumption and heat dissipation in the CPU were 
> more or less proportional but evidently my belief was too naif  :-)
> 
> I guess that the same thing holds for the Athlon64 so prolly centrino is 
> the only CPU were I can get a substantial power saving by decreasing the 
> freq when I don't need much of it (which is almost always), is this 
> correct?


  --  "Frequency Scaling" is when the freqency is increased or decreased,
        usually by fixed factors or to fixed step values.  In an ideal
        world, power consumption is linear with frequency given you keep
        the voltage(s) the same.  In practice, you have leakage, so,
        roughly, at a given set of voltages the eq, looks like:

           Power =  Fixed_Power + Freq * (watt/Hz)

        NOTE: The Fixed_Power and (watt/Hz) values vary based on the
        processor and voltage type.

  --  "Throttling" is a bit like frequency scaling.  The number of clock
        edges is reduced, often by stuttering (times the clock is running
        vs. times it is not), with similar power results.

  --  "Voltage Scaling" roughly changes power as a square with another
        semi-constant:

           Power =  Fixed_Power_2 + V^2 * (watt/V^2)


Now, with all that intro, here's the mechanisms used by the different
chips involved to my knowledge:

  --  Pentium 4 supported Throttling - used to keep the part from
        overheating, and not really useful for power savings in general.

  --  Pentium M, Core, and Core 2 support freq/voltage scaling and maybe (?)
        also Throttling.  Don't know when they started supporting it in
        server configs.

  --  (don't remember what Athlon (K7) supported, I think the mobile
        version had freq/voltage scaling)

  --  Opteron/Athlon64 support freq/voltage scaling across the whole line.


I think the multi-socket (i.e. SMP server) support for freq/voltage
scaling only came into products since about 2004-2005.  I know
Opterons/Athlon64's had it in 2004, but don't know/remember where it
intersected into Intel products.

Results of all this says that you get *less* than linear benefits from
scaling frequency back or throttling alone.  Luckily, you can often
reduce voltage when scaling back the frequency, and in that case it's
really the voltage reduction which gives you most of the power savings.


So, whenever possible use a freq/voltage scaling combination, which is
what Enhanced Speedstep and Powernow have been doing for a while.


--
    Erich Stefan Boleyn     <erich@uruk.org>     http://www.uruk.org/
"Reality is truly stranger than fiction; Probably why fiction is so popular"

Re: do Pentium4 & PentiumM consume about the same at their lowest possible freq?

[Arjan van de Ven]

On Fri, 2007-08-03 at 15:34 -0700, Erich Boleyn wrote:

> Results of all this says that you get *less* than linear benefits from
> scaling frequency back or throttling alone.  Luckily, you can often
> reduce voltage when scaling back the frequency, and in that case it's
> really the voltage reduction which gives you most of the power savings.
> 
> 
> So, whenever possible use a freq/voltage scaling combination, which is
> what Enhanced Speedstep and Powernow have been doing for a while.

the situation is more complex than this though....

The CPU uses a LOT less power when idle. In practice on most cpus
(except when you overclock like crazy and go to high voltages) it's more
power efficient to go full speed to be idle as quickly as possible, so
that you get the power saving from idle.

This is especially the case for things like throttling; where you
basically keep the cpu just at a higher voltage for a longer time than
you would have otherwise...

-- 
if you want to mail me at work (you don't), use arjan (at) linux.intel.com
Test the interaction between Linux and your BIOS via http://www.linuxfirmwarekit.org