Mobile Intel Pentium 4 + ICH5
Hi,
I have a laptop with an Mobile Intel(R) Pentium(R) 4 cpu and a 82801EB ICH5 southbridge.
I can use the p4-clockmod driver, but it doesn't seem to save any power because it only changes the frequency and doesn't do voltage scaling. The speedstep-ich driver only seems to support ICH2, 3 and 4 but not 5. Is voltage scaling speedstep possible for this laptop?
cheers,
Ronald
Quoted 9 lines · Ronald van der Meer
Hi,
I have a laptop with an Mobile Intel(R) Pentium(R) 4 cpu and a 82801EB ICH5 southbridge.
I can use the p4-clockmod driver, but it doesn't seem to save any power because it only changes the frequency and doesn't do voltage scaling. The speedstep-ich driver only seems to support ICH2, 3 and 4 but not 5. Is voltage scaling speedstep possible for this laptop?
Hi Ronald,
The p4-clockmod driver only modulates the CPU-clock. That means that the clock is running like this: 1111000000001111000000001111 (1=on, 0=off).
Because the clock is running at max CPU speed when on, the voltage cannot be lowered. In theory, the voltage could be dropped when the clock is off. But because it takes some time before the voltage stabilises again, that would mean that the modulation frequency should become so slow that it would be very noticable (e.g. jerky mouse movement).
So what you need is speedstep-ich. That driver can both change CPU frequency and CPU voltage. But you say it does not support ICH5. Did you try it already or did you only read the documentation?
BTW. What labtop do you have?
Regards,
Bas.
Hi,
Quoted 8 lines
The p4-clockmod driver only modulates the CPU-clock. That means that the clock is running like this: 1111000000001111000000001111 (1=on, 0=off).
Because the clock is running at max CPU speed when on, the voltage cannot be lowered. In theory, the voltage could be dropped when the clock is off. But because it takes some time before the voltage stabilises again, that would mean that the modulation frequency should become so slow that it would be very noticable (e.g. jerky mouse movement).
Thanx for the info on the p4-clockmod driver. It seems like a pretty much useless cpu-feature to me since it doesn't save any power. It just slows stuff down, which maybe even costs more battery-power per task simply because the task will take longer to finish. If this is true isn't it even misleading for this driver to be in the power-management part of the kernel because this implies it saves power? A warning at least seems very appropriate to me.
Quoted 3 lines
So what you need is speedstep-ich. That driver can both change CPU frequency and CPU voltage. But you say it does not support ICH5. Did you try it already or did you only read the documentation?
I tried a rather old patch (20031203) which didn't work. I checked the current source of the speedstep-ich driver. The function speedstep_detect_chipset in line 186 of speedstep-ich.c doesn't seem to check for my southbridge. This matches what is in the documentation. So trying again seems pretty useless.
So does anybody have any clue about speedstep support with the ICH5 ?
Quoted 1 line
BTW. What labtop do you have?
It's a Medion MD40100, for more info on it's internals check
http://home.conceptsfa.nl/~revdmeer/md40100/
Ronald
Quoted 7 lines · Ronald van der Meer
Hi,
Quoted 1 line
BTW. What labtop do you have?
It's a Medion MD40100, for more info on it's internals check
http://home.conceptsfa.nl/~revdmeer/md40100/
Could you please send 'lspci -n' ?
BTW, I don't see an ICH-5 mobile version on intel site yet ?
Signature · 4 lines
Bruno Ducrot -- Which is worse: ignorance or apathy? -- Don't know. Don't care.
Quoted 10 lines · Bruno Ducrot
Quoted 7 lines · Ronald van der Meer
Hi,
Quoted 1 line
BTW. What labtop do you have?
It's a Medion MD40100, for more info on it's internals check
http://home.conceptsfa.nl/~revdmeer/md40100/Could you please send 'lspci -n' ?
00:00.0 Class 0600: 8086:2570 (rev 02) 00:01.0 Class 0604: 8086:2571 (rev 02) 00:1d.0 Class 0c03: 8086:24d2 (rev 02) 00:1d.1 Class 0c03: 8086:24d4 (rev 02) 00:1d.2 Class 0c03: 8086:24d7 (rev 02) 00:1d.3 Class 0c03: 8086:24de (rev 02) 00:1d.7 Class 0c03: 8086:24dd (rev 02) 00:1e.0 Class 0604: 8086:244e (rev c2) 00:1f.0 Class 0601: 8086:24d0 (rev 02) 00:1f.1 Class 0101: 8086:24db (rev 02) 00:1f.3 Class 0c05: 8086:24d3 (rev 02) 00:1f.5 Class 0401: 8086:24d5 (rev 02) 00:1f.6 Class 0703: 8086:24d6 (rev 02) 01:00.0 Class 0300: 10de:032c (rev a1) 03:04.0 Class 0607: 104c:ac8e 03:04.1 Class 0607: 104c:ac8e 03:04.2 Class 0c00: 104c:802e 03:04.3 Class 0180: 104c:ac8f 03:05.0 Class 0200: 10ec:8139 (rev 10) 03:06.0 Class 0280: 1260:3890 (rev 01)
Quoted 1 line
BTW, I don't see an ICH-5 mobile version on intel site yet ?
I don't think it's a special mobile version. Probably this one:
http://developer.intel.com/design/chipsets/datashts/252523.htm
would this totally rule out the using speedstep which the CPU does seem to support?
Quoted 5 lines · Ronald van der Meer
I don't think it's a special mobile version. Probably this one:
http://developer.intel.com/design/chipsets/datashts/252523.htmwould this totally rule out the using speedstep which the CPU does seem to support?
Unfortunately, it does seem so. The ICH-5 is a desktop version and does not include the special registers necessary for speedstep control.
Does the acpi cpufreq-driver (latest(!) 2.6.) work?
Dominik
_______________________________________________
Cpufreq mailing list
[email protected]
http://www.linux.org.uk/mailman/listinfo/cpufreq
Quoted 9 lines · Dominik Brodowski
Quoted 5 lines · Ronald van der Meer
I don't think it's a special mobile version. Probably this one:
http://developer.intel.com/design/chipsets/datashts/252523.htmwould this totally rule out the using speedstep which the CPU does seem to support?
Unfortunately, it does seem so. The ICH-5 is a desktop version and does not include the special registers necessary for speedstep control.
Indeed. Those registers are 'reserved' as for others non-mobile ICH version. I really don't know what will happens if the OP force detection.
Signature · 4 lines
Bruno Ducrot -- Which is worse: ignorance or apathy? -- Don't know. Don't care.
Quoted 9 lines · Dominik Brodowski
Quoted 5 lines · Ronald van der Meer
I don't think it's a special mobile version. Probably this one:
http://developer.intel.com/design/chipsets/datashts/252523.htmwould this totally rule out the using speedstep which the CPU does seem to support?
Unfortunately, it does seem so. The ICH-5 is a desktop version and does not include the special registers necessary for speedstep control.
What registers would be required for the Intel Mobile Pentium 4 to do it's speedstep thing?
Reading through it's datasheet
(http://www.intel.com/design/mobile/datashts/253028.htm)
I couldn't find any reference to southbridges or it's registers. It seems to require significantly less signals from outside than older intel mobile procs. For example:
The processor determines it's own voltage by driving it's VID[4:0] pins (table 3, page 15 of datasheet) to a Voltage Regulation Module (VRM). So this processor doesn't need a mobile ICH and an external latch (as shown in the block diagram figure 5-14, on page 155 of the ICH4-M datasheet) to provide this functionality.
The datasheet btw doesn't talk about speedstep, but about enhanced speedstep. The VID[4:0] signal driving described above seem like one of the enhancements to me, anybody know any others ?
Infact the way I read the datasheet all it takes to do speedstepping is:
Switch to max-performance mode : Set GHI# pin low, put processor into Deep Sleep state, return to normal state.
Switch to battery optimized mode : Set GHI# pin high, put processor into Deep Sleep state, return to normal state.
This raises 2 questions:
Is changing the processors state to "Deep Sleep" and back possible in linux?
To what is the processors GHI# pin connected?
I realise all this research may turn out to be pointless, but I would like to be 100% certain if speedstepping is possible or not. Just saying 'the ICH5 is not a special mobile version so speedstepping is not possible' seems too easy to me.
Quoted 1 line
Does the acpi cpufreq-driver (latest(!) 2.6.) work?
I'm afraid it will not because 'cat /proc/acpi/processor/CPU/performance' always showed '<not supported>' on the kernels I tried. But next time I try a recent 2.6 kernel I'll try.
Ronald
Quoted 11 lines · Ronald van der Meer
Quoted 9 lines · Dominik Brodowski
Quoted 5 lines · Ronald van der Meer
I don't think it's a special mobile version. Probably this one:
http://developer.intel.com/design/chipsets/datashts/252523.htmwould this totally rule out the using speedstep which the CPU does seem to support?
Unfortunately, it does seem so. The ICH-5 is a desktop version and does not include the special registers necessary for speedstep control.
Quoted 8 lines
What registers would be required for the Intel Mobile Pentium 4 to do it's speedstep thing?
Reading through it's datasheet
(http://www.intel.com/design/mobile/datashts/253028.htm)
I couldn't find any reference to southbridges or it's registers. It seems to require significantly less signals from outside than older intel mobile procs. For example:
Normal, that's under NDA, mostly, apart the register associated to the ICH?m southbridge.
Quoted 5 lines
The processor determines it's own voltage by driving it's VID[4:0] pins (table 3, page 15 of datasheet) to a Voltage Regulation Module (VRM). So this processor doesn't need a mobile ICH and an external latch (as shown in the block diagram figure 5-14, on page 155 of the ICH4-M datasheet) to provide this functionality.
You program the VID output via the VRM module which is under NDA, and actualy I don't know if that is correct :)
Quoted 3 lines
The datasheet btw doesn't talk about speedstep, but about enhanced speedstep. The VID[4:0] signal driving described above seem like one of the enhancements to me, anybody know any others ?
No. That's how it work back to pentium III (coppermine, speedstep capable).
Quoted 6 lines
Infact the way I read the datasheet all it takes to do speedstepping is:
Switch to max-performance mode : Set GHI# pin low, put processor into Deep Sleep state, return to normal state.
Switch to battery optimized mode : Set GHI# pin high, put processor into Deep Sleep state, return to normal state.
You toggle the GHI# pin via an enhanced propritary ASIC for which you will not get specification from Intel unless you sign a NDA.
Quoted 5 lines
This raises 2 questions:
Is changing the processors state to "Deep Sleep" and back possible in linux?
To what is the processors GHI# pin connected?
To a propritary ASIC already mentionned that you will find no reference from Intel litterature.
Quoted 4 lines
I realise all this research may turn out to be pointless, but I would like to be 100% certain if speedstepping is possible or not. Just saying 'the ICH5 is not a special mobile version so speedstepping is not possible' seems too easy to me.
No. That's the truth unfortunately. All Speedstep capable processor (apart perhaps the Banias on centrino platform) need at least:
A mobile VRM,
A propritary ASIC,
A northbridge which will connected somehow to that ASIC,
A southbridge, that one integrated a part of the state machine associated to SpeedStep technology (the trigger part, mostly).
It was possible, back to PIII, to have a southbridge that do not incorporate a part of the state machine associated to speedstep, but then it was necessary to reserve a GPO and a GPI for speedstep.
It is possible after all that it is still the case when the southbridge is not a mobile version, but frankly, I doubt that is the case. To be sure, you may try to dump the GPO associated when you boot on AC, then on Battery, and see if that make a difference, like it's done for the PIIX4 IO version.
Cheers,
Signature · 4 lines
Bruno Ducrot -- Which is worse: ignorance or apathy? -- Don't know. Don't care.
Quoted 7 lines · Ronald van der Meer
Thanx for the info on the p4-clockmod driver. It seems like a pretty much useless cpu-feature to me since it doesn't save any power. It just slows stuff down, which maybe even costs more battery-power per task simply because the task will take longer to finish. If this is true isn't it even misleading for this driver to be in the power-management part of the kernel because this implies it saves power? A warning at least seems very appropriate to me.
On systems where no "idle" states C2 or C3 are offered, the p4-clockmod driver can save (some, not much) power, AFAIK. However, for P4-M or other advanced mobile processors, it is more or less useless. Because of this, a warning was added a couple of days ago to the 2.6. p4-clockmod driver stating:
printk(KERN_DEBUG PFX "Warning: Pentium 4-M detected. The speedstep-ich or acpi cpufreq \n");
printk(KERN_DEBUG PFX "modules offers voltage scaling in addition of frequency scaling. You\n");
printk(KERN_DEBUG PFX "should use either one instead of p4-clockmod, if possible.\n");or
printk(KERN_DEBUG PFX "Warning: Pentium M detected. The speedstep_centrino module\n");
printk(KERN_DEBUG PFX "offers voltage scaling in addition of frequency scaling. You\n");
printk(KERN_DEBUG PFX "should use that instead of p4-clockmod, if possible.\n");
Dominik_______________________________________________
Cpufreq mailing list
[email protected]
http://www.linux.org.uk/mailman/listinfo/cpufreq