Hi,

I had posted a message earlier, but its missing now, sorry for the repost.

I'm using a phenom 9500 under Linux, but unable to scale to any frequency except the minimum (1.1Ghz) and the maximum (2.2Ghz). Under Windows XP I can scale to the intermediate frequencies, but not in Linux.

I have cpufreq installed, and a pretty recent linux kernel 2.6.23 or later, the k8 powernow driver seems to have loaded ok. Is it possible these intermediate values are missing from where cpufreq would be reading them from? (some P-state table that linux is not reading?)

Anything I should check?

Thanks,

Try this:

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies

This will tell you what frequencies cpufreq recognizes. If it just sees the minimum and maximum frequencies, then you may need to wait for a new version of powernow to ship.

-------------------------

Thanks for the response.

I've tried this and it only lists the two (min and max freqs).

Is there someway to figure out if its my setup thats at fault or if I can just wait for an update?

I guess in the meantime I'll just wait on this for an update to cpufreq.

If it works with Windows, then your processor and motherboard support the correct FID/VID mappings. And since you can get some scaling with the chip in Linux, this means that cpufreq is set up properly. Just wait for an updated cpufreq and the problem should go away.

-------------------------

FYI, for each transition, the voltage alone has at least 20us of setting time, and setting the frequency won't be trivial either.
In the Phenom Bios and OS writer's guide, it says that it's pretty common to have the machine spend 100's of us in a power-state transition.

Slamming from the min to the max isn't all that terrible. If you're concerned with power savings, just set the governor to be conversative, or userspace, and lock the cores to the min, and release them only as necessary. However, I don't think the kernel scheduler are aware of the different corespeed and would schedule accordingly.

There is no plan to ship a new version of PowerNow! at this time.

If Windows is giving you more pstates than Linux is, that usually means there is something odd about the ACPI objects and Linux isn't interpreting them correctly. This usually requires an update to the Linux ACPI code to resolve, not to the PowerNow! driver.