Undervolting with kernel 2.6 [Done!]

widan · Posted: Sun Mar 05, 2006 2:42 pm Post subject:

The BIOS developer's guide says you should not have two low-fid entries:

Dio-Sama · n00b Joined: 18 Mar 2005 Posts: 19

so all in all the worst that could happen when adding more steps is a little whacky behaviour like crashes at some of the transitions...
i'm fine with that, so i guess i'll just wait for morfic to update his patch then.
_________________
"Death is but perchance to dream."

Erdie · Posted: Wed Mar 15, 2006 9:01 am Post subject:

I just tried out this hack by manually changing the kernel sources. This is really a good hint. I just saved about 10 degrees of temperature under heavy load. I am now running a 1.350 V@2GHz. I made a mistake at first and accidentally my machine runs a 1.1V@2GHz

But I could recompile my kernel modules without crash and now it is running save with 1.35V. This is really "cool", because the fan power management was broken since a certain time and i am running now the heatpipe fan at constant RPM with 5V. With this hack it will be not risky even at summer temperatures.

Erdie
_________________
Desktop AMD Ryzen 9 5900X 32GB RAM, Asus GF GTX 1060.
Notebook Tuxedo Pulse 15 Gen1 AMD Ryzen 7 4800H mit Radeon Vega 7
Raspberry Pi 1 + 2 + 3B+ + Zero W

superwutze · Posted: Wed Mar 15, 2006 3:24 pm Post subject:

just ran some tests on my machine and wanted to share the results:

my system: tyan tiger k8w with 2 opteron 246 cpus, 2gb ram, ....

at first i patched a 2.6.15-r7 kernel and it borked. after some testing i realized that my system crashed the instant it changed frequencies. later i realized that it crashed only when i changed the voltage of the slowest mode. so i did the following:
original -> now
2ghz/1500mv -> 2ghz/1300mv
1.8ghz/1400mv -> 1.8ghz/1250mv
1ghz/1200mv -> 1ghz/1200mv
full load: 40/44 deg -> 34/37 deg (cpu0/cpu1)

and everything works fine! :lol:

then i simply copied the powernow-k8.c from 2.6.15 to 2.6.14 because i can't use 2.6.15. it kills my cordless desktop set, pptp tunnels don't work anymore and (most important) i can't watch tv anymore!!

of course i diffed the two at first, but the differences are minimal.

now i have a running system at 2.6.14, undervolting works perfect with ondemand governor, vpn with pptp and mppe works again but i still can't watch tv

that worked last with 2.6.14-r3.

so long and happy undervolting,
andy
_________________
bill who? micro what?

Elmo234 · n00b Joined: 08 Oct 2004 Posts: 48

I just started coding a program that is used as a daemon and undervolts the athlon64 cpu's.

I write the new vid's using /dev/cpu/*/msr

You can have a look at it here:

http://tuxamito.com.es/cpupw/

tnt · Veteran Joined: 27 Feb 2004 Posts: 1227

freke · Posted: Thu Mar 16, 2006 9:52 pm Post subject:

No doubt that an ebuild would be nice, but I think I'd rather have morfic sneak this one into the gentoo-sources...

It seems to be working flawlessly? At least I haven't seen any reports on it having broken anything so come on morfic, try and sneak it in

mark · Tux's lil' helper Joined: 04 Jun 2002 Posts: 119

superwutze · Posted: Sun Mar 19, 2006 7:41 pm Post subject:

i use bios 3.02 since the day it was released and it solved a lot of different problems. the board was always stable with windows, but performance was another thing. i like to max my computers out so windows is just for some real special compatibility problems for some customers. but for quite some time there was (alas!) no linux for me with that board. that was because of the agp design of the 8511 chip from amd. there is a huge compatibility problem with that agp-chip and a lot of power-demanding graphic-cards. back then i used a radeon 9800xt (till today the topmost energy-demanding card of all). since i switched to a x700pro and flashed to 3.02 everything simply works. my problems were mostly with my addon-cards. the board itself is (except for the poor memory performance with just one memory bus) a very nice and fine thing. it's a pity that there will be no dual-core support.

i always had a sata harddisk with this board, but my cdrw and dvdrw drives are pata. there were never any problems.

if you have some specific questions just ask!

so long,
andy
_________________
bill who? micro what?

mark · Tux's lil' helper Joined: 04 Jun 2002 Posts: 119

Well I have had some agp probs, a change in memory helped but I have agp disabled at the moment.

I have recently upgraded to 2x248 opterons and wanted to enable powernow but when I scale down to minimum I get a hard lock.

You're lock up problems sounded very similar.

We run the same hios so thats not it.

I have been in contact with the maintainor of powernow-k8 and it was suggested I had a libata oops, but you are using sata with I presume libata so that should not be it either.

I'm thinking of playing with scaling without a voltage drop but that seems a long shot.

I'm just trying to figure out where to go from here realy.

Thanks for coming back to me

Mark
_________________
Regards

Mark

Elmo234 · n00b Joined: 08 Oct 2004 Posts: 48

brihall · Posted: Sun Mar 19, 2006 11:07 pm Post subject:

I tried cpupw. Worked fine with my CPU, an Opteron 165 (two C&Q steps) on an Asrock 939Dual-Sata2.

# cpupw -s
System has 1 physical CPU's
CPU 0-> voltage: 1.150000 v, VID: 16, frequency: 1000 Mhz, FID: 2

I did have one small disappointment, though. I was hoping to increase voltage over the max provided by my BIOS settings with this. No luck, I am still limited to 1.40V (default +.05). Using a kernel patch (powernow-k8-vcore_list-2.6.16-rc2.diff) to allow me to select the C&Q voltages also doesn't allow >1.40V. Not sure if this is a hardware limitation (I know you can modify the motherboard for up to 1.55V by shorting VID pins), but it would be nice if cpupw could override this limit.

I remember using the first version to set the current CPU voltage from the command line, but now I can't remember what the syntax was. Could you add that info to the --help listing?

Also, since I am overclocking, the frequency listed is incorrect. Perhaps the current frequency could be output (from /proc/cpuinfo) as well as the FID.

Elmo234 · n00b Joined: 08 Oct 2004 Posts: 48

If modifying the kernel and this program didn't work it might be because a hardware limitation. In my laptop I have the full range of voltage but in my desktop the voltage will never go under 1.1v and VID will never decrease 18. But I haven't have time to investigate what causes this.

I just realised that I didn't program any way to just set a different voltage from the command line. I will to it very soon, I promise! (probably version 0.1.2)

About the speed and voltage... I just calculate it from the FID and VID. For K8 I always suppose a 100 Mhz speed bus. I don't know how to guess the FSB (or HT link) speed... (any ideas) Las option will be calculate it form /proc/cpuinfo

Thanks for your information

widan · Posted: Mon Mar 20, 2006 12:13 am Post subject:

Elmo234 · n00b Joined: 08 Oct 2004 Posts: 48

brihall · Posted: Mon Mar 20, 2006 1:01 am Post subject:

a64-vid seems to indicate that 1.55V is available, but trying to get there fails. The attempt to do this at the command line that I referred to in the previous post was using a64-vid, not cpupw.

# a64-vid
Maximum freq. : 0x0a (1800 MHz)
Startup freq. : 0x0a (1800 MHz)
Current freq. : 0x02 (1000 MHz)

Voltage range : 0x00 (1550 mV) -> 0x06 (1400 mV)
Startup voltage : 0x08 (1350 mV)
Current voltage : 0x10 (1150 mV

# a64-vid 1550
Maximum freq. : 0x0a (1800 MHz)
Startup freq. : 0x0a (1800 MHz)
Current freq. : 0x02 (1000 MHz)

Voltage range : 0x00 (1550 mV) -> 0x06 (1400 mV)
Startup voltage : 0x08 (1350 mV)
Current voltage : 0x10 (1150 mV)

Target voltage : 0x00 (1550 mV)
New control MSR : 0x0000000100010002

New voltage : 0x06 (1400 mV) [FAILURE]

widan · Posted: Mon Mar 20, 2006 8:34 am Post subject:

Devport · Guru Joined: 15 Dec 2004 Posts: 361

brihall · Posted: Tue Mar 21, 2006 2:52 pm Post subject:

The powernow patch works for me on 2.6.15 and 2.6.16, just booted 2.6.16 this morning:

Linux syrinx 2.6.16-ck1 #1 SMP PREEMPT Mon Mar 20 15:10:50 MST 2006 x86_64 Dual Core AMD Opteron(tm) Processor 165 GNU/Linux

powernow-k8: Found 2 AMD Athlon 64 / Opteron processors (version 1.60.0)
powernow-k8: Requested Voltage for id0 is 1350 mV default voltage is 1350
powernow-k8: Performing required adjustment of 0 mV or 0 25mV steps
powernow-k8: Requested Voltage for id1 is 1075 mV default voltage is 1300
powernow-k8: Performing required adjustment of 225 mV or 9 25mV steps
powernow-k8: 0 : fid 0xa (1800 MHz), vid 0x8 (1350 mV)
powernow-k8: 1 : fid 0x2 (1000 MHz), vid 0x13 (1075 mV)
cpu_init done, current fid 0xa, vid 0x8

I do it via the kernel boot line, I remember being successful doing when inserting the module. From my grub.conf:

kernel /boot/bzImage vga=normal noexec=on noexec32=on root=/dev/sda2 ro elevator=cfq powernow-k8.vcore_list=1375,1100

I have also noticed that the order of the list may vary, depending on the processor. For example, on my S754 3000+, this works:

kernel /boot/bzImage vga=normal noexec=on noexec32=on root=/dev/hda3 ro powernow-k8.vcore_list=900,1100,1200

Some experimentation required...

If it could be cleaned up a little so the parameter order is consistent, it might even be allowed into the main kernel.

Devport · Guru Joined: 15 Dec 2004 Posts: 361

mark · Tux's lil' helper Joined: 04 Jun 2002 Posts: 119

I've just tried the patch and discovered that the lockups on my tyan k8w (2875) disappear if I set the lowest voltage to 1300 rather than the default 1200. Whilst I have not tested extensively 1300 works at all frequencies for my dual opteron 248's.

Now I had thought that cpufreq scaled the cpu's in tandem but this no longer seems to be true. Frequencies scale independently. However if I manually set both cpu's to the same frequency, 1200 and maybe lower seems to be doable.

I have no idea how to automatically scale them in tandem, or even if that is worth it. It may be better to let them scale inbependently at 1300mv.

So I don't have all the answers, I have no idea if this is a bug or a hardware issue, but the info may help others at some point.

Note that even without this patch default voltages hang my system when scaled independently on all recent kernels

Mark
_________________
Regards

Mark

squash · Tux's lil' helper Joined: 22 Jun 2002 Posts: 79

Neither a64-vid nor cpupw work for me, however I am having luck with the powernow-k8.c patch.

obrut<- · Posted: Thu Apr 13, 2006 2:19 pm Post subject:

on my system powernow_k8 does not load ("mp systems not supported by psb bios structure") but i'd like to clock down my cpu when idle. the powernow driver for windows does not load either. but i can change fid and vid using rightmark's rmclock. so there should be a way make it work under linux. perhaps a patch for this module. i can test the voltages when running windows, so i would be satisfied when i can define the voltages in a kernel source file (as i do it with the speedstep_centrino module on my notebook). but i don't know how to change the powernow_k8 module

Voluspa · n00b Joined: 20 Mar 2006 Posts: 3 Location: Sweden