Thinkpad T400 / P8600

[Bzzz]

Hi,
a little success story after many drawbacks...
I'm not a very experienced Linux user, I switched over from (15+ years of ) Windows just a few months ago. Of course I had some Ubuntu VMs, some dual boot configs, I fixed some issues on other machines and I've been using it on public PCs, but that's only scratching on the surface and not looking what's under the bonnet. So my new main system, the T400, is now a Linux-only machine. It is equipped with a Core 2 Duo P8600, 2x2,4 GHz. I'm using Kubuntu 10.10 with kernel 2.6.35-25-generic.

First of all, well, installing the PPA kernel. Pretty easy, adding it to repositories, updating, installing, rebooting, everything fine.

Installing phc-intel: A nightmare.

1. phc-intel-0.3.2-12-1 contains two kernel folders, 2.6.33 and 2.6.34. Faking the directory name helps a little, but I don't know if there are any side-effects.
2. the make process doesn't work. After testing the commands in the Makefile, I found out that the package "byobu" is needed to provide "shell". This may be fine for GNOME users, but KDE/XFCE users may not have byobu pre-installed. No word on that in the help files. No error message when running the script!
3. Makefile also fails, because "KERNELVERSION=$(shell awk -F\" '/REL/ {print $$2}' $(UTSREL))" yields "2.6.35.10" (on the 2.6.35-25-generic-phc kernel!). Manually setting the value to the correct one: One step further.
4. KERNELSRC points to the generic folder, don't know if that is an issue, but I also corrected that to "=/lib/modules/2.6.35-25-generic-phc/build"
5. Even now the script cannot copy the .ko file to the right folder. It just echoes the four lines of the install: section. I once had a flawless run, but at that time it was still pointing to the 2.6.35-10 folder, which had no impact. At that point, I learned that copying stuff from this folder to the current phc one is a bad BAD BAD idea

Hmm...somehow I got it working. Now the PHCtool works, but it cannot throttle the CPU or fix it to some specific frequency for testing lower voltage/frequency pairs other than the maximum. I also don't know how to do this without phc.

Moreover, PHCtool shows me a pair 2401 MHz/ VID 39 and 2400 MHz / VID 34. I got both stable* on 31, but I don't know what's the reason for this extra MHz (needing extra 62,5 mV).
*Prime yields a perfect run over 10 hours (19,x CPU hours) just to crash one of the workers when I got up this morning using the machine again, running on VID 30. I don't know if that is related to some swapping / I/O problems I have (only 2 GB of RAM installed atm). Prime with "blend" settings nearly freezes the machine, maybe due to memory consumption and therefore endless swapping.

Well, on VID 31 instead of 39/34, I have a peak consumption of about 34.2 W compared to 36.4 W. Not that much to make any difference, but still promising for a undervolting setup at the 800 MHz / VID 17 point for idling. Also, I never had a consumption above 23 W in any real-life scenario.
Please let me know if you have any ideas how to bench with pinned frequency on maximum CPU load...

[Docet]

If you want a proper guide to clean install PHC and PHCtool take a look here:

http://linuxsolver.blogspot.com/2011/02 ... buntu.html

Anyway the benchmark/autoset VIDs at frequency is still missing, there is an obsolete script which I tested and it is not usable. Maybe I'm going to develop a new version of it in the months to come.

For what concerns the mysterious 1MHz step (I have a T9800 box and, yes I have that last, final and useless step too), I read somewhere it should be a sort of "turbo mode"... don't know how and why it works, bu it looks like it needs much more voltage than the close -1MHz step, so I suggest you a differential setting... I use a -0.1V (looking at default values) at any freq step... try to find your best ^_^


Cheers