Re: Speedstep on Celeron SU2300

Re: Speedstep on Celeron SU2300

[Tiago Marques]

Hi,
I'm having a problem with this processor not having frequency steps
and apparently only voltage steps. I find it very strange but that's
what Intel's documentation suggests. I can't load acpi-cpufreq because
it doesn't find any device and battery life in linux is suffering
around 20% less due to this. I have confirmed that the CPU supports
Speedstep, just this very strange variation.
Can I somehow help with this to get it fixed? Who are the current maintainers?
Thanks,
Tiago

Speedstep on Celeron SU2300

[Tiago Marques]

Hi,

Can someone give me some hints on this?
Best regards

On Sun, Aug 29, 2010 at 5:54 AM, Tiago Marques <tiagomnm@gmail.com> wrote:
>
> Hi,
> I'm having a problem with this processor not having frequency steps
> and apparently only voltage steps. I find it very strange but that's
> what Intel's documentation suggests. I can't load acpi-cpufreq because
> it doesn't find any device and battery life in linux is suffering
> around 20% less due to this. I have confirmed that the CPU supports
> Speedstep, just this very strange variation.
> Can I somehow help with this to get it fixed? Who are the current maintainers?
> Thanks,
> Tiago

Re: Speedstep on Celeron SU2300

[Thomas Renninger]

On Sunday 29 August 2010 06:54:44 Tiago Marques wrote:
> Hi,
> I'm having a problem with this processor not having frequency steps
> and apparently only voltage steps. I find it very strange but that's
> what Intel's documentation suggests. I can't load acpi-cpufreq because
> it doesn't find any device and battery life in linux is suffering
> around 20% less due to this.
Where do you have the 20% info from, I doubt you verified it?

> I have confirmed that the CPU supports
> Speedstep, just this very strange variation.
You are not the only one:
https://bugzilla.kernel.org/show_bug.cgi?id=16072
[HP Pavilion dm1-1110ev] Cpufreq doesn't work at all ( Intel Celeron U2300 )
or the last comments (or search for U2300) here:
http://software.intel.com/en-us/articles/enhanced-intel-speedstepr-technology-and-demand-based-switching-on-linux/

> Can I somehow help with this to get it fixed? Who are the current maintainers?
Not exporting cpu frequencies seem to be intended for this cpu for whatever reasons.
If you have efficient C-states, frequency states are not that important.
You could try msr-tools to access frequency switching MSRs directly.
But you need to know the supported frequencies. Intel generally publishes
such documentation for all of their released CPUs.

You could then try to set the frequency by:
FREQUENCY="max/min_freq_from Intel spec"
modprobe msr
wrmsr --processor X 0x199 $FREQUENCY

This is not a solution, but you could figure out whether switching works
at all and if, you could test whether you really have 20% more of battery life time,
but I expect you won't get that much.

      Thomas

Re: Speedstep on Celeron SU2300

[Tiago Marques]

Hi Thomas. Thanks for the message.

On Wed, Sep 1, 2010 at 11:40 PM, Thomas Renninger <trenn@suse.de> wrote:
> On Sunday 29 August 2010 06:54:44 Tiago Marques wrote:
>> Hi,
>> I'm having a problem with this processor not having frequency steps
>> and apparently only voltage steps. I find it very strange but that's
>> what Intel's documentation suggests. I can't load acpi-cpufreq because
>> it doesn't find any device and battery life in linux is suffering
>> around 20% less due to this.
> Where do you have the 20% info from, I doubt you verified it?

Yes. Since it seems no one can't return Windows licenses for refunds
anymore, I have went ahead and booted windows on it, without any
driver installed and just configured it to have frequency scaling
working, which in this case is only voltage scaling.
I measured almost 6 hours of battery life and the processor & chipset
frequently had the fan stop when idling.

In linux, in the same conditions, I got less than 4 hours and the fan
never stops. I configured a PCI-express power saving feature on the
kernel and it seems to have dropped noise a bit. Battery life is still
not great and the fan still never stops. I'm trying to find something
with which I can measure the actual power going through the AC adapter
but for now battery life tests is pretty much all I can do.
The 20% figure is obviously optimistic since I'm accounting a positive
effect from the PCI-Express power saving feature to not quote 33%
less.

>
>> I have confirmed that the CPU supports
>> Speedstep, just this very strange variation.
> You are not the only one:
> https://bugzilla.kernel.org/show_bug.cgi?id=16072
> [HP Pavilion dm1-1110ev] Cpufreq doesn't work at all ( Intel Celeron U2300 )
> or the last comments (or search for U2300) here:
> http://software.intel.com/en-us/articles/enhanced-intel-speedstepr-technology-and-demand-based-switching-on-linux/
>
>> Can I somehow help with this to get it fixed? Who are the current maintainers?
> Not exporting cpu frequencies seem to be intended for this cpu for whatever reasons.

It doesn't have more than one. I checked in windows and I also can't
change frequency, despite reports from a person on the contrary:

http://scottiestech.info/2010/02/05/how-to-increase-your-laptops-battery-life-a-lot-with-crystalcpuid/

I have confirmed that the GM45 chipset doesn't support frequency
scaling of the FSB on Intel's datasheet, so the speed he reports of
98MHz is a bug. I found similar frequencies upon use. Since I
performed my tests with LCD brightness in full, that would account for
the one hour difference on the best result, while the less than four
hours he also reports may be due to speedstep not working before he
messed with configurations.

I tried the same program and the CPU is locked at 6x, although the
voltage isn't, it's somewhere from 0.925 to 1.075v if I recall
correctly.

> If you have efficient C-states, frequency states are not that important.

I know. But what about voltage? Intel's datasheet clearly states this
processor supports two voltage states, I guess for the purpose of
relaxing clock binning requirements for these CPUs. I'm thinking it
does scale when in windows.
The Pentium SU4100 has similar problems but he allows one p-state,
from 1300MHz to 1200MHz.
I can't seem to find the lowest multiplier available on these
platforms, CrystalCPUID lists mine as 6x, hence 1200MHz. I thought
Core 2's could idle at something like 800MHz and I find it strange
that this one can't also.

> You could try msr-tools to access frequency switching MSRs directly.
> But you need to know the supported frequencies. Intel generally publishes
> such documentation for all of their released CPUs.
>
> You could then try to set the frequency by:
> FREQUENCY="max/min_freq_from Intel spec"
> modprobe msr
> wrmsr --processor X 0x199 $FREQUENCY
>
> This is not a solution, but you could figure out whether switching works
> at all and if, you could test whether you really have 20% more of battery life time,
> but I expect you won't get that much.

That probably won't work, given my tests in windows. Can I do that for
the voltage also? I also tried the linux-phc modules for debugging but
it also doesn't load.

Best regards,
Tiago

>
>      Thomas
>

Re: Speedstep on Celeron SU2300

[Thomas Renninger]

On Thursday 02 September 2010 01:17:14 Tiago Marques wrote:
> Hi Thomas. Thanks for the message.
> 
> On Wed, Sep 1, 2010 at 11:40 PM, Thomas Renninger <trenn@suse.de> wrote:
> > On Sunday 29 August 2010 06:54:44 Tiago Marques wrote:
> >> Hi,
> >> I'm having a problem with this processor not having frequency steps
> >> and apparently only voltage steps. I find it very strange but that's
> >> what Intel's documentation suggests. I can't load acpi-cpufreq because
> >> it doesn't find any device and battery life in linux is suffering
> >> around 20% less due to this.
> > Where do you have the 20% info from, I doubt you verified it?
> 
> Yes. Since it seems no one can't return Windows licenses for refunds
> anymore, I have went ahead and booted windows on it, without any
> driver installed and just configured it to have frequency scaling
> working, which in this case is only voltage scaling.
> I measured almost 6 hours of battery life and the processor & chipset
> frequently had the fan stop when idling.
> 
> In linux, in the same conditions, I got less than 4 hours and the fan
> never stops. I configured a PCI-express power saving feature on the
> kernel and it seems to have dropped noise a bit. Battery life is still
> not great and the fan still never stops. I'm trying to find something
> with which I can measure the actual power going through the AC adapter
> but for now battery life tests is pretty much all I can do.
There is current battery power drain somehwere in
/proc/acpi/battery/*/*
It normally updates not that often, but may be better and accurate enough if you
take several values, than waiting for the battery got drained.

> The 20% figure is obviously optimistic since I'm accounting a positive
> effect from the PCI-Express power saving feature to not quote 33%
> less.
Yep, this could be the graphics card as well.
What kind of graphics card has it and which driver are you using?

> >> I have confirmed that the CPU supports
> >> Speedstep, just this very strange variation.
> > You are not the only one:
> > https://bugzilla.kernel.org/show_bug.cgi?id=16072
> > [HP Pavilion dm1-1110ev] Cpufreq doesn't work at all ( Intel Celeron U2300 )
> > or the last comments (or search for U2300) here:
> > http://software.intel.com/en-us/articles/enhanced-intel-speedstepr-technology-and-demand-based-switching-on-linux/
> >
> >> Can I somehow help with this to get it fixed? Who are the current maintainers?
> > Not exporting cpu frequencies seem to be intended for this cpu for whatever reasons.
> 
> It doesn't have more than one. I checked in windows and I also can't
> change frequency, despite reports from a person on the contrary:
> 
> http://scottiestech.info/2010/02/05/how-to-increase-your-laptops-battery-life-a-lot-with-crystalcpuid/
> 
> I have confirmed that the GM45 chipset doesn't support frequency
> scaling of the FSB on Intel's datasheet, so the speed he reports of
> 98MHz is a bug. I found similar frequencies upon use. Since I
> performed my tests with LCD brightness in full, that would account for
> the one hour difference on the best result, while the less than four
> hours he also reports may be due to speedstep not working before he
> messed with configurations.
> 
> I tried the same program and the CPU is locked at 6x, although the
> voltage isn't, it's somewhere from 0.925 to 1.075v if I recall
> correctly.
> 
> > If you have efficient C-states, frequency states are not that important.
> 
> I know. But what about voltage? Intel's datasheet clearly states this
> processor supports two voltage states, I guess for the purpose of
> relaxing clock binning requirements for these CPUs. I'm thinking it
> does scale when in windows.
> The Pentium SU4100 has similar problems but he allows one p-state,
> from 1300MHz to 1200MHz.
> I can't seem to find the lowest multiplier available on these
> platforms, CrystalCPUID lists mine as 6x, hence 1200MHz. I thought
> Core 2's could idle at something like 800MHz and I find it strange
> that this one can't also.
Interesting. Possibly you have luck finding a document from Intel about
this CPU describing this a bit. I'd be interested in the outcome, but don't
have to time to dig for it.

     Thomas

Re: Speedstep on Celeron SU2300

[Tiago Marques]

On Thu, Sep 2, 2010 at 12:52 AM, Thomas Renninger <trenn@suse.de> wrote:
> On Thursday 02 September 2010 01:17:14 Tiago Marques wrote:
>> Hi Thomas. Thanks for the message.
>>
>> On Wed, Sep 1, 2010 at 11:40 PM, Thomas Renninger <trenn@suse.de> wrote:
>> > On Sunday 29 August 2010 06:54:44 Tiago Marques wrote:
>> >> Hi,
>> >> I'm having a problem with this processor not having frequency steps
>> >> and apparently only voltage steps. I find it very strange but that's
>> >> what Intel's documentation suggests. I can't load acpi-cpufreq because
>> >> it doesn't find any device and battery life in linux is suffering
>> >> around 20% less due to this.
>> > Where do you have the 20% info from, I doubt you verified it?
>>
>> Yes. Since it seems no one can't return Windows licenses for refunds
>> anymore, I have went ahead and booted windows on it, without any
>> driver installed and just configured it to have frequency scaling
>> working, which in this case is only voltage scaling.
>> I measured almost 6 hours of battery life and the processor & chipset
>> frequently had the fan stop when idling.
>>
>> In linux, in the same conditions, I got less than 4 hours and the fan
>> never stops. I configured a PCI-express power saving feature on the
>> kernel and it seems to have dropped noise a bit. Battery life is still
>> not great and the fan still never stops. I'm trying to find something
>> with which I can measure the actual power going through the AC adapter
>> but for now battery life tests is pretty much all I can do.
> There is current battery power drain somehwere in
> /proc/acpi/battery/*/*
> It normally updates not that often, but may be better and accurate enough if you
> take several values, than waiting for the battery got drained.

Thanks, I'll look into that for kernel tuning. I just can't use to
compare with windows because I don't think I have similar information
there.

>
>> The 20% figure is obviously optimistic since I'm accounting a positive
>> effect from the PCI-Express power saving feature to not quote 33%
>> less.
> Yep, this could be the graphics card as well.
> What kind of graphics card has it and which driver are you using?
>
>> >> I have confirmed that the CPU supports
>> >> Speedstep, just this very strange variation.
>> > You are not the only one:
>> > https://bugzilla.kernel.org/show_bug.cgi?id=16072
>> > [HP Pavilion dm1-1110ev] Cpufreq doesn't work at all ( Intel Celeron U2300 )
>> > or the last comments (or search for U2300) here:
>> > http://software.intel.com/en-us/articles/enhanced-intel-speedstepr-technology-and-demand-based-switching-on-linux/
>> >
>> >> Can I somehow help with this to get it fixed? Who are the current maintainers?
>> > Not exporting cpu frequencies seem to be intended for this cpu for whatever reasons.
>>
>> It doesn't have more than one. I checked in windows and I also can't
>> change frequency, despite reports from a person on the contrary:
>>
>> http://scottiestech.info/2010/02/05/how-to-increase-your-laptops-battery-life-a-lot-with-crystalcpuid/
>>
>> I have confirmed that the GM45 chipset doesn't support frequency
>> scaling of the FSB on Intel's datasheet, so the speed he reports of
>> 98MHz is a bug. I found similar frequencies upon use. Since I
>> performed my tests with LCD brightness in full, that would account for
>> the one hour difference on the best result, while the less than four
>> hours he also reports may be due to speedstep not working before he
>> messed with configurations.
>>
>> I tried the same program and the CPU is locked at 6x, although the
>> voltage isn't, it's somewhere from 0.925 to 1.075v if I recall
>> correctly.
>>
>> > If you have efficient C-states, frequency states are not that important.
>>
>> I know. But what about voltage? Intel's datasheet clearly states this
>> processor supports two voltage states, I guess for the purpose of
>> relaxing clock binning requirements for these CPUs. I'm thinking it
>> does scale when in windows.
>> The Pentium SU4100 has similar problems but he allows one p-state,
>> from 1300MHz to 1200MHz.
>> I can't seem to find the lowest multiplier available on these
>> platforms, CrystalCPUID lists mine as 6x, hence 1200MHz. I thought
>> Core 2's could idle at something like 800MHz and I find it strange
>> that this one can't also.
> Interesting. Possibly you have luck finding a document from Intel about
> this CPU describing this a bit. I'd be interested in the outcome, but don't
> have to time to dig for it.

http://www.intel.com/design/mobile/datashts/321111.pdf

Please have a look at this document, page 30, note 10, at the top of the page.

Do you have any experience with SU processors? I have no idea of idle
clock of the Core 2 SU CPUs, perhaps it is also 1200MHz, hence the
limitation on idle clock? Could be the lowest multiplier available.

Best regards

>
>     Thomas
>

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Thomas Renninger]

Hi,

I am adding lesswatts.org and linux-pm list.
I expect you found out why you do not get frequency/P- states and it
seem to be correct. On these lists, people can help you further
to find out Linux vs Windows battery drain differences.

If you have the same backlight settings, I expect C-state or graphics
card must be the reason. There is nothing else than CPU or GPU
that drains so much energy for being the reason of
20% more battery life time.
Which graphics card and driver do you use (for nvidia/ati, trying the
binary one for comparison, might show a big difference on a recent
card)?
Which C-state driver do you use (there is an acpi and intel_idle one
with latest kernels):
cat /sys/devices/system/cpu/cpuidle/current_driver
I hope to be able to provide a c-state tool soon, for now you have to
go through:
cat /sys/devices/system/cpu/cpu*/cpuidle/state*
to check which and how often/efficient C-states are used.

Some background (and a comment in the end):

On Saturday 04 September 2010 13:33:26 Tiago Marques wrote:
> On Thu, Sep 2, 2010 at 12:52 AM, Thomas Renninger <trenn-l3A5Bk7waGM@public.gmane.org> wrote:
> > On Thursday 02 September 2010 01:17:14 Tiago Marques wrote:
> >> Hi Thomas. Thanks for the message.
> >>
> >> On Wed, Sep 1, 2010 at 11:40 PM, Thomas Renninger <trenn-l3A5Bk7waGM@public.gmane.org> wrote:
> >> > On Sunday 29 August 2010 06:54:44 Tiago Marques wrote:
> >> >> Hi,
> >> >> I'm having a problem with this processor not having frequency steps
> >> >> and apparently only voltage steps. I find it very strange but that's
> >> >> what Intel's documentation suggests. I can't load acpi-cpufreq because
> >> >> it doesn't find any device and battery life in linux is suffering
> >> >> around 20% less due to this.
> >> > Where do you have the 20% info from, I doubt you verified it?
> >>
> >> Yes. Since it seems no one can't return Windows licenses for refunds
> >> anymore, I have went ahead and booted windows on it, without any
> >> driver installed and just configured it to have frequency scaling
> >> working, which in this case is only voltage scaling.
> >> I measured almost 6 hours of battery life and the processor & chipset
> >> frequently had the fan stop when idling.
> >>
> >> In linux, in the same conditions, I got less than 4 hours and the fan
> >> never stops. I configured a PCI-express power saving feature on the
> >> kernel and it seems to have dropped noise a bit. Battery life is still
> >> not great and the fan still never stops. I'm trying to find something
> >> with which I can measure the actual power going through the AC adapter
> >> but for now battery life tests is pretty much all I can do.
> > There is current battery power drain somehwere in
> > /proc/acpi/battery/*/*
> > It normally updates not that often, but may be better and accurate enough if you
> > take several values, than waiting for the battery got drained.
> 
> Thanks, I'll look into that for kernel tuning. I just can't use to
> compare with windows because I don't think I have similar information
> there.
Best is you get or buy a cheap measure device, in which you can plug
your AC adapter in and remove the battery.
Some electricity company borrow one for free, at least in Germany.
It needs not to be very accurate, you are hunting far above 1Watt drain
differences... Could cost about 10-20 Euro and you can measure your
fridge and other things as well for your overall monthly electricity bill.

... <cut out frequency discussion, see below> ...
> >> I can't seem to find the lowest multiplier available on these
> >> platforms, CrystalCPUID lists mine as 6x, hence 1200MHz. I thought
> >> Core 2's could idle at something like 800MHz and I find it strange
> >> that this one can't also.
> > Interesting. Possibly you have luck finding a document from Intel about
> > this CPU describing this a bit. I'd be interested in the outcome, but don't
> > have to time to dig for it.
> 
> http://www.intel.com/design/mobile/datashts/321111.pdf
> 
> Please have a look at this document, page 30, note 10, at the top of the page.
> 
> Do you have any experience with SU processors?
Not really, but other reports of SU processors without P-states and this info
from the paper you point to:
"The Highest Frequency Mode (HFM) and Lowest Frequency Mode (LFM) refer to the highest
and lowest core operating frequencies supported on the Genuine Intel Processor."
...
"10. SU2300 processor operates at same core frequency in HFM and LFM."

explain the lack of P-states.
> I have no idea of idle
> clock of the Core 2 SU CPUs, perhaps it is also 1200MHz, hence the
> limitation on idle clock? Could be the lowest multiplier available.
I expect lowering power on this kind of CPU is all about C-states.
Possibly these need some special treatment, no idea.
Hopefully one of the Intel guys jump into this thread...

        Thomas

Re: [Discuss] Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Arjan van de Ven]

  On 9/6/2010 1:35 AM, Thomas Renninger wrote:
> Hi,
>
> I am adding lesswatts.org and linux-pm list.
> I expect you found out why you do not get frequency/P- states and it
> seem to be correct. On these lists, people can help you further
> to find out Linux vs Windows battery drain differences.
>
> If you have the same backlight settings, I expect C-state or graphics
> card must be the reason. There is nothing else than CPU or GPU
> that drains so much energy for being the reason of
> 20% more battery life time.
> Which graphics card and driver do you use (for nvidia/ati, trying the
> binary one for comparison, might show a big difference on a recent
> card)?
> Which C-state driver do you use (there is an acpi and intel_idle one
> with latest kernels):
> cat /sys/devices/system/cpu/cpuidle/current_driver
> I hope to be able to provide a c-state tool soon, for now you have to
> go through:
> cat /sys/devices/system/cpu/cpu*/cpuidle/state*
> to check which and how often/efficient C-states are used.

powertop will tell you exactly this

(and the next version will provide much much more detail)

Re: [Discuss] Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

On Mon, Sep 6, 2010 at 4:47 PM, Arjan van de Ven <arjan@linux.intel.com> wrote:
>  On 9/6/2010 1:35 AM, Thomas Renninger wrote:
>>
>> Hi,
>>
>> I am adding lesswatts.org and linux-pm list.
>> I expect you found out why you do not get frequency/P- states and it
>> seem to be correct. On these lists, people can help you further
>> to find out Linux vs Windows battery drain differences.
>>
>> If you have the same backlight settings, I expect C-state or graphics
>> card must be the reason. There is nothing else than CPU or GPU
>> that drains so much energy for being the reason of
>> 20% more battery life time.
>> Which graphics card and driver do you use (for nvidia/ati, trying the
>> binary one for comparison, might show a big difference on a recent
>> card)?
>> Which C-state driver do you use (there is an acpi and intel_idle one
>> with latest kernels):
>> cat /sys/devices/system/cpu/cpuidle/current_driver
>> I hope to be able to provide a c-state tool soon, for now you have to
>> go through:
>> cat /sys/devices/system/cpu/cpu*/cpuidle/state*
>> to check which and how often/efficient C-states are used.
>
> powertop will tell you exactly this
>

Thank you. Please see my previous e-mail. C0, C1 and C4. Do you know
of any way to force the intermediate ones for debugging?

Best regards,
Tiago

> (and the next version will provide much much more detail)
>
>
>

Re: [Discuss] Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Arjan van de Ven]

  On 9/6/2010 3:03 PM, Tiago Marques wrote:
> On Mon, Sep 6, 2010 at 4:47 PM, Arjan van de Ven<arjan@linux.intel.com>  wrote:
>>   On 9/6/2010 1:35 AM, Thomas Renninger wrote:
>>> Hi,
>>>
>>> I am adding lesswatts.org and linux-pm list.
>>> I expect you found out why you do not get frequency/P- states and it
>>> seem to be correct. On these lists, people can help you further
>>> to find out Linux vs Windows battery drain differences.
>>>
>>> If you have the same backlight settings, I expect C-state or graphics
>>> card must be the reason. There is nothing else than CPU or GPU
>>> that drains so much energy for being the reason of
>>> 20% more battery life time.
>>> Which graphics card and driver do you use (for nvidia/ati, trying the
>>> binary one for comparison, might show a big difference on a recent
>>> card)?
>>> Which C-state driver do you use (there is an acpi and intel_idle one
>>> with latest kernels):
>>> cat /sys/devices/system/cpu/cpuidle/current_driver
>>> I hope to be able to provide a c-state tool soon, for now you have to
>>> go through:
>>> cat /sys/devices/system/cpu/cpu*/cpuidle/state*
>>> to check which and how often/efficient C-states are used.
>> powertop will tell you exactly this
>>
> Thank you. Please see my previous e-mail. C0, C1 and C4. Do you know
> of any way to force the intermediate ones for debugging?

intel_idle driver is the only way that's reasonable in amount of work.

for anything else you depend on the grace of the bios.

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

[-- Attachment #1.1: Type: text/plain, Size: 1769 bytes --]

I see. Thanks. It seems that intel_idle isn't supported on Core 2 based
Celerons.

Best regards

On Tue, Sep 7, 2010 at 12:28 AM, Arjan van de Ven <arjan-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>wrote:

>  On 9/6/2010 3:03 PM, Tiago Marques wrote:
>
>> On Mon, Sep 6, 2010 at 4:47 PM, Arjan van de Ven<arjan-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
>>  wrote:
>>
>>>  On 9/6/2010 1:35 AM, Thomas Renninger wrote:
>>>
>>>> Hi,
>>>>
>>>> I am adding lesswatts.org and linux-pm list.
>>>> I expect you found out why you do not get frequency/P- states and it
>>>> seem to be correct. On these lists, people can help you further
>>>> to find out Linux vs Windows battery drain differences.
>>>>
>>>> If you have the same backlight settings, I expect C-state or graphics
>>>> card must be the reason. There is nothing else than CPU or GPU
>>>> that drains so much energy for being the reason of
>>>> 20% more battery life time.
>>>> Which graphics card and driver do you use (for nvidia/ati, trying the
>>>> binary one for comparison, might show a big difference on a recent
>>>> card)?
>>>> Which C-state driver do you use (there is an acpi and intel_idle one
>>>> with latest kernels):
>>>> cat /sys/devices/system/cpu/cpuidle/current_driver
>>>> I hope to be able to provide a c-state tool soon, for now you have to
>>>> go through:
>>>> cat /sys/devices/system/cpu/cpu*/cpuidle/state*
>>>> to check which and how often/efficient C-states are used.
>>>>
>>> powertop will tell you exactly this
>>>
>>>  Thank you. Please see my previous e-mail. C0, C1 and C4. Do you know
>> of any way to force the intermediate ones for debugging?
>>
>
> intel_idle driver is the only way that's reasonable in amount of work.
>
> for anything else you depend on the grace of the bios.
>
>

[-- Attachment #1.2: Type: text/html, Size: 2689 bytes --]

[-- Attachment #2: Type: text/plain, Size: 161 bytes --]

_______________________________________________
Discuss mailing list
Discuss-ovCJLiDH4KkgsBAKwltoeQ@public.gmane.org
http://lists.lesswatts.org/listinfo/discuss

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

I see. Thanks. It seems that intel_idle isn't supported on Core 2
based Celerons.
Best regards



> On Tue, Sep 7, 2010 at 12:28 AM, Arjan van de Ven <arjan@linux.intel.com>
> wrote:
>>
>>  On 9/6/2010 3:03 PM, Tiago Marques wrote:
>>>
>>> On Mon, Sep 6, 2010 at 4:47 PM, Arjan van de Ven<arjan@linux.intel.com>
>>>  wrote:
>>>>
>>>>  On 9/6/2010 1:35 AM, Thomas Renninger wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>> I am adding lesswatts.org and linux-pm list.
>>>>> I expect you found out why you do not get frequency/P- states and it
>>>>> seem to be correct. On these lists, people can help you further
>>>>> to find out Linux vs Windows battery drain differences.
>>>>>
>>>>> If you have the same backlight settings, I expect C-state or graphics
>>>>> card must be the reason. There is nothing else than CPU or GPU
>>>>> that drains so much energy for being the reason of
>>>>> 20% more battery life time.
>>>>> Which graphics card and driver do you use (for nvidia/ati, trying the
>>>>> binary one for comparison, might show a big difference on a recent
>>>>> card)?
>>>>> Which C-state driver do you use (there is an acpi and intel_idle one
>>>>> with latest kernels):
>>>>> cat /sys/devices/system/cpu/cpuidle/current_driver
>>>>> I hope to be able to provide a c-state tool soon, for now you have to
>>>>> go through:
>>>>> cat /sys/devices/system/cpu/cpu*/cpuidle/state*
>>>>> to check which and how often/efficient C-states are used.
>>>>
>>>> powertop will tell you exactly this
>>>>
>>> Thank you. Please see my previous e-mail. C0, C1 and C4. Do you know
>>> of any way to force the intermediate ones for debugging?
>>
>> intel_idle driver is the only way that's reasonable in amount of work.
>>
>> for anything else you depend on the grace of the bios.
>>
>
>
_______________________________________________
Discuss mailing list
Discuss@lesswatts.org
http://lists.lesswatts.org/listinfo/discuss

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

On Mon, Sep 6, 2010 at 9:35 AM, Thomas Renninger <trenn@suse.de> wrote:
> Hi,
>
> I am adding lesswatts.org and linux-pm list.
> I expect you found out why you do not get frequency/P- states and it
> seem to be correct. On these lists, people can help you further
> to find out Linux vs Windows battery drain differences.

Thank you very much.

>
> If you have the same backlight settings, I expect C-state or graphics
> card must be the reason. There is nothing else than CPU or GPU
> that drains so much energy for being the reason of
> 20% more battery life time.
> Which graphics card and driver do you use (for nvidia/ati, trying the
> binary one for comparison, might show a big difference on a recent
> card)?

Intel's, the most recent driver in Gentoo's portage, which is  2.12.0

> Which C-state driver do you use (there is an acpi and intel_idle one
> with latest kernels):
> cat /sys/devices/system/cpu/cpuidle/current_driver
> I hope to be able to provide a c-state tool soon, for now you have to
> go through:
> cat /sys/devices/system/cpu/cpu*/cpuidle/state*
> to check which and how often/efficient C-states are used.
>

C states seem ok...ish. I get C0, C1 and C4, though not anything in
between. Don't know how to check in windows and usually it shouldn't
in leaps, right? That might be a problem. I'm getting around 75% in
C4, the rest almost exclusively in C0. Perhaps the transitions are the
problem? I think I previously saw normal rates of closer to 98% so
I'll have to check what's going now. Too many wakeups from i915, usb
and wlan.
Powertop shows that the BIOS reports C1 and C4 support, hence only
these C-stats.

> Some background (and a comment in the end):
>
> On Saturday 04 September 2010 13:33:26 Tiago Marques wrote:
>> On Thu, Sep 2, 2010 at 12:52 AM, Thomas Renninger <trenn@suse.de> wrote:
>> > On Thursday 02 September 2010 01:17:14 Tiago Marques wrote:
>> >> Hi Thomas. Thanks for the message.
>> >>
>> >> On Wed, Sep 1, 2010 at 11:40 PM, Thomas Renninger <trenn@suse.de> wrote:
>> >> > On Sunday 29 August 2010 06:54:44 Tiago Marques wrote:
>> >> >> Hi,
>> >> >> I'm having a problem with this processor not having frequency steps
>> >> >> and apparently only voltage steps. I find it very strange but that's
>> >> >> what Intel's documentation suggests. I can't load acpi-cpufreq because
>> >> >> it doesn't find any device and battery life in linux is suffering
>> >> >> around 20% less due to this.
>> >> > Where do you have the 20% info from, I doubt you verified it?
>> >>
>> >> Yes. Since it seems no one can't return Windows licenses for refunds
>> >> anymore, I have went ahead and booted windows on it, without any
>> >> driver installed and just configured it to have frequency scaling
>> >> working, which in this case is only voltage scaling.
>> >> I measured almost 6 hours of battery life and the processor & chipset
>> >> frequently had the fan stop when idling.
>> >>
>> >> In linux, in the same conditions, I got less than 4 hours and the fan
>> >> never stops. I configured a PCI-express power saving feature on the
>> >> kernel and it seems to have dropped noise a bit. Battery life is still
>> >> not great and the fan still never stops. I'm trying to find something
>> >> with which I can measure the actual power going through the AC adapter
>> >> but for now battery life tests is pretty much all I can do.
>> > There is current battery power drain somehwere in
>> > /proc/acpi/battery/*/*
>> > It normally updates not that often, but may be better and accurate enough if you
>> > take several values, than waiting for the battery got drained.
>>
>> Thanks, I'll look into that for kernel tuning. I just can't use to
>> compare with windows because I don't think I have similar information
>> there.
> Best is you get or buy a cheap measure device, in which you can plug
> your AC adapter in and remove the battery.
> Some electricity company borrow one for free, at least in Germany.
> It needs not to be very accurate, you are hunting far above 1Watt drain
> differences... Could cost about 10-20 Euro and you can measure your
> fridge and other things as well for your overall monthly electricity bill.

I have one, cost me around 35eur. Unfortunately it's precision seems
to be around 5W, so It will be quite hard to check it from there. Will
give it a try though, had been thinking of that.

>
> ... <cut out frequency discussion, see below> ...
>> >> I can't seem to find the lowest multiplier available on these
>> >> platforms, CrystalCPUID lists mine as 6x, hence 1200MHz. I thought
>> >> Core 2's could idle at something like 800MHz and I find it strange
>> >> that this one can't also.
>> > Interesting. Possibly you have luck finding a document from Intel about
>> > this CPU describing this a bit. I'd be interested in the outcome, but don't
>> > have to time to dig for it.
>>
>> http://www.intel.com/design/mobile/datashts/321111.pdf
>>
>> Please have a look at this document, page 30, note 10, at the top of the page.
>>
>> Do you have any experience with SU processors?
> Not really, but other reports of SU processors without P-states and this info
> from the paper you point to:
> "The Highest Frequency Mode (HFM) and Lowest Frequency Mode (LFM) refer to the highest
> and lowest core operating frequencies supported on the Genuine Intel Processor."
> ...
> "10. SU2300 processor operates at same core frequency in HFM and LFM."
>
> explain the lack of P-states.

Yes, but shouldn't it detect P-States with different voltages, as in
windows? (even as stupid as this sounds) I sure loved to get it always
stuck at the lower clock, given that I usually perform maintenance on
PCs I highly doubt the fan would get dust clogged to the point of
taking the CPU to temperatures that would cause problems.
This looks like it doesn't support Speedstep even though Intel says it
does but the fact is that Windows can work with two voltages, which
always improves power.

>> I have no idea of idle
>> clock of the Core 2 SU CPUs, perhaps it is also 1200MHz, hence the
>> limitation on idle clock? Could be the lowest multiplier available.
> I expect lowering power on this kind of CPU is all about C-states.
> Possibly these need some special treatment, no idea.
> Hopefully one of the Intel guys jump into this thread...

I'd love to hear from them, because C-States isn't Speedstep.

Best regards,
Tiago

>
>        Thomas
>

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Thomas Renninger]

On Monday 06 September 2010 23:47:58 Tiago Marques wrote:
> On Mon, Sep 6, 2010 at 9:35 AM, Thomas Renninger <trenn@suse.de> wrote:
..
> C states seem ok...ish. I get C0, C1 and C4, though not anything in
> between. Don't know how to check in windows and usually it shouldn't
> in leaps, right? That might be a problem. I'm getting around 75% in
> C4, the rest almost exclusively in C0. Perhaps the transitions are the
> problem? I think I previously saw normal rates of closer to 98% so
> I'll have to check what's going now. Too many wakeups from i915, usb
> and wlan.
> Powertop shows that the BIOS reports C1 and C4 support, hence only
> these C-stats.
Yep, I'd concentrate on this first. If idle, the system should be in deepest
C-state (nearly) all of the time. Perfect tool for this is powertop.
You could also boot into runlevel S. If the system is much more idle
there you could load drivers one by one to find out whether there is a bad
one keeping your cpu busy/awake.

C-states also should "rule-out" P-states. If you are in C4 it shouldn't matter
whether your freq is up or down, the core's freq and voltage should be
ramped down to a minimum on latest processors (I've never measured that,
but this is what I've heard from Intel guys themselves...).

Good luck,

    Thomas

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

Hello,

On Tue, Sep 7, 2010 at 12:47 PM, Thomas Renninger <trenn@suse.de> wrote:
>
> On Monday 06 September 2010 23:47:58 Tiago Marques wrote:
> > On Mon, Sep 6, 2010 at 9:35 AM, Thomas Renninger <trenn@suse.de> wrote:
> ..
> > C states seem ok...ish. I get C0, C1 and C4, though not anything in
> > between. Don't know how to check in windows and usually it shouldn't
> > in leaps, right? That might be a problem. I'm getting around 75% in
> > C4, the rest almost exclusively in C0. Perhaps the transitions are the
> > problem? I think I previously saw normal rates of closer to 98% so
> > I'll have to check what's going now. Too many wakeups from i915, usb
> > and wlan.
> > Powertop shows that the BIOS reports C1 and C4 support, hence only
> > these C-stats.
> Yep, I'd concentrate on this first. If idle, the system should be in deepest
> C-state (nearly) all of the time. Perfect tool for this is powertop.
> You could also boot into runlevel S. If the system is much more idle
> there you could load drivers one by one to find out whether there is a bad
> one keeping your cpu busy/awake.
>

The guilty party was the broadcom binary driver, it was waking the CPU
up 30k times per second. I've tried with an USB wlan card and it
hovers around 70-20 wakeups per second. Does that look normal?

>
> C-states also should "rule-out" P-states. If you are in C4 it shouldn't matter
> whether your freq is up or down, the core's freq and voltage should be
> ramped down to a minimum on latest processors (I've never measured that,
> but this is what I've heard from Intel guys themselves...).
>

I'm still due to actually measure times again but from casual
observation the CPU fan is still on considerably more time than in
Windows. It seems to me that the lower voltage in idle that Windows
can use does impact temperatures a bit, after all the CPU does have to
wake from time to time.
I'll be measuring this directly from the CPU voltage regulator when I
find the time, as to leave out any doubts on what's really going on.

Best regards,
Tiago

> Good luck,
>
>    Thomas

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Matthew Garrett]

On Mon, Sep 06, 2010 at 10:47:58PM +0100, Tiago Marques wrote:

> Yes, but shouldn't it detect P-States with different voltages, as in
> windows? (even as stupid as this sounds) I sure loved to get it always
> stuck at the lower clock, given that I usually perform maintenance on
> PCs I highly doubt the fan would get dust clogged to the point of
> taking the CPU to temperatures that would cause problems.
> This looks like it doesn't support Speedstep even though Intel says it
> does but the fact is that Windows can work with two voltages, which
> always improves power.

If there's only one frequency then there's no reason to have multiple 
voltages - the voltage will already be at the minimum possible for the 
core to be stable at that frequency. Entering C4 will typically result 
in the voltage dropping as parts of the core are disabled.

Does your chip have the "est" flag in /proc/cpuinfo? If not, it doesn't 
support speedstep.

-- 
Matthew Garrett | mjg59@srcf.ucam.org

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Thomas Renninger]

On Tuesday 07 September 2010 17:22:46 Matthew Garrett wrote:
> On Mon, Sep 06, 2010 at 10:47:58PM +0100, Tiago Marques wrote:
> 
> > Yes, but shouldn't it detect P-States with different voltages, as in
> > windows? (even as stupid as this sounds) I sure loved to get it always
> > stuck at the lower clock, given that I usually perform maintenance on
> > PCs I highly doubt the fan would get dust clogged to the point of
> > taking the CPU to temperatures that would cause problems.
> > This looks like it doesn't support Speedstep even though Intel says it
> > does but the fact is that Windows can work with two voltages, which
> > always improves power.
> 
> If there's only one frequency then there's no reason to have multiple 
> voltages - the voltage will already be at the minimum possible for the 
> core to be stable at that frequency. Entering C4 will typically result 
> in the voltage dropping as parts of the core are disabled.
> 
> Does your chip have the "est" flag in /proc/cpuinfo? If not, it doesn't 
> support speedstep.
That is somewhat strange. These CPUs show est support in cpuinfo, also
compare with (not only about HP, but also other OEMs):
[Bug 16072] [HP Pavilion dm1-1110ev] Cpufreq doesn't work at all ( Intel Celeron U2300 )
https://bugzilla.kernel.org/show_bug.cgi?id=16072

But Tiago digged out an Intel spec which says that the SU processors do only
run at one freq:
http://www.intel.com/design/mobile/datashts/321111.pdf
page 30, note 10, at the top of the page.

I could imagine cpuid should not export est capabilities for these and
cpufreq drivers should not complain that an est capable CPU is found for
which no frequencies get exported, but probably only Intel can tell for
sure.

     Thomas

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

On Tue, Sep 7, 2010 at 5:43 PM, Thomas Renninger <trenn@suse.de> wrote:
> On Tuesday 07 September 2010 17:22:46 Matthew Garrett wrote:
>> On Mon, Sep 06, 2010 at 10:47:58PM +0100, Tiago Marques wrote:
>>
>> > Yes, but shouldn't it detect P-States with different voltages, as in
>> > windows? (even as stupid as this sounds) I sure loved to get it always
>> > stuck at the lower clock, given that I usually perform maintenance on
>> > PCs I highly doubt the fan would get dust clogged to the point of
>> > taking the CPU to temperatures that would cause problems.
>> > This looks like it doesn't support Speedstep even though Intel says it
>> > does but the fact is that Windows can work with two voltages, which
>> > always improves power.
>>
>> If there's only one frequency then there's no reason to have multiple
>> voltages - the voltage will already be at the minimum possible for the
>> core to be stable at that frequency. Entering C4 will typically result
>> in the voltage dropping as parts of the core are disabled.
>>
>> Does your chip have the "est" flag in /proc/cpuinfo? If not, it doesn't
>> support speedstep.
> That is somewhat strange. These CPUs show est support in cpuinfo, also
> compare with (not only about HP, but also other OEMs):
> [Bug 16072] [HP Pavilion dm1-1110ev] Cpufreq doesn't work at all ( Intel Celeron U2300 )
> https://bugzilla.kernel.org/show_bug.cgi?id=16072
>
> But Tiago digged out an Intel spec which says that the SU processors do only
> run at one freq:
> http://www.intel.com/design/mobile/datashts/321111.pdf
> page 30, note 10, at the top of the page.
>
> I could imagine cpuid should not export est capabilities for these and
> cpufreq drivers should not complain that an est capable CPU is found for
> which no frequencies get exported, but probably only Intel can tell for
> sure.

I'd love input on this, because with the lower voltage being there I
could probably get away with using it even in load usage since I tend
to keep vents clean and temperatures here aren't high. coretemp is
showing me around 60ºC in load. To me, having a lower voltage when the
CPU has to come out of C4 into C0 is an advantage, one which I hope to
measure. When having the CPU run at C0 in light loads(<50%) would also
be useful to have the lower voltage there, kicking it up when the two
cores are +80% to keep stability in check with the temperature rise(if
needed...). Or am I completely wrong and when having 50% load the CPU
also does a lot of C4 idling?

The only reason I took the plunge with a Celeron processor, besides
being the only one available in the DM1 around here, was that it
supported EIST. I would not get one laptop without it, so let's say I
was a bit perplexed when I tried to get it to run and failed.

Best regards,
Tiago


>
>     Thomas
>
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

On Tue, Sep 7, 2010 at 4:22 PM, Matthew Garrett <mjg59@srcf.ucam.org> wrote:
> On Mon, Sep 06, 2010 at 10:47:58PM +0100, Tiago Marques wrote:
>
>> Yes, but shouldn't it detect P-States with different voltages, as in
>> windows? (even as stupid as this sounds) I sure loved to get it always
>> stuck at the lower clock, given that I usually perform maintenance on
>> PCs I highly doubt the fan would get dust clogged to the point of
>> taking the CPU to temperatures that would cause problems.
>> This looks like it doesn't support Speedstep even though Intel says it
>> does but the fact is that Windows can work with two voltages, which
>> always improves power.
>
> If there's only one frequency then there's no reason to have multiple
> voltages - the voltage will already be at the minimum possible for the
> core to be stable at that frequency.

Unfortunately, that's not what the Intel document says. It's beyond my
understanding why anyone would lock these CPUs also at 1200MHz and
claim them to have Speedstep, but then again I'm not the one trying to
pull out quarter after quarter of record profits while delivering
misleading products.

I think they're supplying a lower voltage to improve battery life and
a higher one to control processor stability when the laptop gets in a
state where it can reach close to 100ºC and the lower voltage might
not be enough - like when the fan ducts get stuck with dust or in a
very hot day/country/region. It doesn't make any sense to keep the
clock out of the equation but we also had to wait for years(and the
Atom) for laptops to have C4 state enabled in the BIOS. IIRC, three
years ago one laptop that made it to C2 was uncommon, let alone
deeper.

It seems the designers of this line were not in their right mind, as
they have also done things like enabling VT-x in these Celeron
processors while the more expensive Pentium SU4100 doesn't have VT-x
and also has only 1200MHz and 1300MHz as P-states. I still can't find
if the other Core 2 CULV processors can go below 1200MHz and a proper
reasoning for that to be so.

> Entering C4 will typically result
> in the voltage dropping as parts of the core are disabled.
>
> Does your chip have the "est" flag in /proc/cpuinfo? If not, it doesn't
> support speedstep.

Yes. I have to confirm this with a multimeter but as I can see in
Windows, it seems that it uses both voltages with 1200MHz in the two
P-states, while the acpi-cpufreq module fails because it can't find
more than onde differente clock in the p-states the BIOS exposes.

Best regards,
Tiago Marques

>
> --
> Matthew Garrett | mjg59@srcf.ucam.org
>
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Len Brown]

> I'm getting around 75% in C4, the rest almost exclusively in C0.

This 25% in C0 is likely all of your battery life problem.

use top and powertop to find out what is running and why,
and see if it is something you can prevent from running.

If your CPU is more than 1% buys when "profoundly idle",
then it has a significant software or configuration problem.

cheers,
Len Brown, Intel Open Source Technology Center

ps. note for comparisons...  your XP partition should have a
"perfmon" application that can draw a pretty graph of cpu usage,
including adding counters for the various C-states.

Re: Speedstep on Celeron SU2300 - 20% more battery lifetime on Windows

[Tiago Marques]

Hello,

On Tue, Sep 7, 2010 at 5:20 PM, Len Brown <lenb@kernel.org> wrote:
>> I'm getting around 75% in C4, the rest almost exclusively in C0.
>
> This 25% in C0 is likely all of your battery life problem.
>
> use top and powertop to find out what is running and why,
> and see if it is something you can prevent from running.
>
> If your CPU is more than 1% buys when "profoundly idle",
> then it has a significant software or configuration problem.

As mentioned in another e-mail, it was a problem with the broadcom
driver. I have an atheros 9285 card on order, let's see how that one
holds, the USB one did ok.

> cheers,
> Len Brown, Intel Open Source Technology Center
>
> ps. note for comparisons...  your XP partition should have a
> "perfmon" application that can draw a pretty graph of cpu usage,
> including adding counters for the various C-states.
>

Using Server 2008 for tests, currently broken. Alternatives?

Best regards,
Tiago