How Can I disable CPUscalling in Fedora 16

[kelvinhbo]

This is really annoying for my system, I used to disabled it before F16 by stopping the cpuspeed service, but they dropped that in F16, so is there any other way to stop Fedora from scalling down my CPU to save power?

[stevea]

I think ....

for f in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor ; do echo performance > $f ; done

[sexycatsinhats]

You can edit /etc/sysconfig/cpupower and make sure it says "performance" instead of "on_demand" after CPUPOWER_ on all entries for a more permanent fix (stays on bootup). Also, check your frequency with

cpupower frequency-info

to verify if it's disabled.

[DBelton]

unfortunately, cpufreq and cpupower don't always give the correct information.

From log at boot:

[    0.000000] hpet clockevent registered
[    0.000000] Fast TSC calibration using PIT
[    0.000000] Detected 2257.445 MHz processor.

from cpupower frequency-info:

[Me@tower20 ~]$ cpupower frequency-info
analyzing CPU 0:
  driver: p4-clockmod
  CPUs which run at the same hardware frequency: 0
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 10.00 ms.
  hardware limits: 225 MHz - 1.80 GHz
  available frequency steps: 225 MHz, 450 MHz, 675 MHz, 900 MHz, 1.13 GHz, 1.35 GHz, 1.58 GHz, 1.80 GHz
  available cpufreq governors: conservative, userspace, powersave, ondemand, performance
  current policy: frequency should be within 225 MHz and 1.80 GHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency is 1.80 GHz.
  boost state support:
    Supported: yes
    Active: yes

as you can see, the kernel saw it correctly, 2257.445 MHz, but cpufreq will not allow setting the maximum frequency higher than what it detects the processor to be, which is 1800 MHz.

Edit:
I disabled EIST (intel speedstep) in the BIOS to prevent cpufreq from scaling it back. Using speedstep and cpufreq with the performance governor, I couldn't set the maximum frequency higher than 1800 MHz, so I disabled EIST and it just loads the p4-clockmod governor.

[RupertPupkin]

I've just noticed the same problem on my F16 system. I had some cli apps that were running about twice as slow as they used to under F14 and F15 on the same hardware, and I couldn't figure out why. Then I saw this:

$ cat /proc/cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 15
model           : 55
model name      : AMD Athlon(tm) 64 Processor 3700+
stepping        : 2
cpu MHz         : 1000.000
cache size      : 1024 KB
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt lm 3dnowext 3dnow up rep_good nopl pni lahf_lm
bogomips        : 1999.50
TLB size        : 1024 4K pages
clflush size    : 64
cache_alignment : 64
address sizes   : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp

So my 2.2GHz cpu is running at only 1GHz. WTF?!?! No wonder my stuff was running twice as slow.
The kernel properly detects my cpu:

$ dmesg|grep -i mhz
[    0.000000] Detected 2199.455 MHz processor.
[    2.164979] powernow-k8: fid 0x2 (1000 MHz), vid 0x12
[    2.188101] powernow-k8: fid 0xa (1800 MHz), vid 0xa
[    2.210863] powernow-k8: fid 0xc (2000 MHz), vid 0x8
[    2.233272] powernow-k8: fid 0xe (2200 MHz), vid 0x8

I installed the kernel-tools package and found this:

$ cpupower frequency-info
analyzing CPU 0:
  driver: powernow-k8
  CPUs which run at the same hardware frequency: 0
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 1.24 ms.
  hardware limits: 1000 MHz - 2.20 GHz
  available frequency steps: 1000 MHz, 1.80 GHz, 2.00 GHz, 2.20 GHz
  available cpufreq governors: conservative, userspace, powersave, ondemand, performance
  current policy: frequency should be within 1000 MHz and 1000 MHz.
                  The governor "performance" may decide which speed to use
                  within this range.
  current CPU frequency is 1000 MHz.
  boost state support:
    Supported: no
    Active: no

So even though I should be getting 2.2GHz, the "current policy" (set by someone who isn't me) restricts my cpu to a "range" of 1GHz to 1GHz! Again, WTF?!?! I don't see a way of changing this, which is what's really ticking me off. The fedoraproject.org docs on this are completely outdated, referring only to the old setup on F14. My bet is this change was made to stop laptops from overheating with the 3.x kernels (which apparently is a problem), but not giving users a way to change this is inexcusable.

[PabloTwo]

I think if you're not using the "ondemand" governor, your missing the boat.

BASH:~/-> cpupower frequency-info
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0 1 2 3
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency: 10.0 us.
  hardware limits: 1.60 GHz - 3.30 GHz
  available frequency steps: 3.30 GHz, 3.30 GHz, 3.10 GHz, 2.90 GHz, 2.70 GHz, 2.50 GHz, 2.30 GHz, 2.10 GHz, 1.90 GHz, 1.70 GHz, 1.60 GHz
  available cpufreq governors: conservative, userspace, powersave, ondemand, performance
  current policy: frequency should be within 1.60 GHz and 3.30 GHz.
                  The governor "ondemand" may decide which speed to use
                  within this range.
  current CPU frequency is 1.60 GHz.
  boost state support:
    Supported: yes
    Active: yes
    25500 MHz max turbo 4 active cores
    25500 MHz max turbo 3 active cores
    25500 MHz max turbo 2 active cores
    25500 MHz max turbo 1 active cores

Looks like something is amiss with your "performance" governor".

---------- Post added at 12:44 PM ---------- Previous post was at 12:39 PM ----------

You might want to check these on your system also:

BASH:~/-> cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_min_freq 
1600000
BASH:~/-> cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq
3301000
BASH:~/-> cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
1600000
BASH:~/-> cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
3301000

---------- Post added at 12:49 PM ---------- Previous post was at 12:44 PM ----------

And this:

BASH:~/-> cat /sys/devices/system/cpu/cpu0/cpufreq/bios_limit 
3301000

[RupertPupkin]

Thanks, Pablo. Here's what I get:

$ cat /sys/devices/system/cpu/cpu0/cpufreq/bios_limit
2200000
$ cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_min_freq
1000000
$ cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq
2200000
$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_min_freq
1000000
$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1000000

Looks like the scaling_max_freq is wrong. I tried the ondemand governor before and it didn't help. What ticks me off is that there used to be a cpuspeed service, which I always disabled to get the maximum cpu frequency. But it looks like F16 puts that into the kernel now, and you have to modify the cpu behavior with software tools that don't appear to always work properly. The bad part of all this is that users' systems may run slower than they should.

[PabloTwo]

Looks like you just need to: echo 2200000 /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
and do that for each core [cpu0 cpu1 cpu2 cpu3]
There could be something(s) else configured out of whack also... just poke around in /sys/devices/system/cpu/ and look for anything else that looks to be not set correctly.

[RupertPupkin]

I tried "echo 2200000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq" as root and it didn't work:

# echo 2200000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq
1000000

Looks like that scaling_max_freq is getting hard-coded and can't be changed. I have a bad feeling that it may be related to the available frequencies being displayed in different units (1000 MHz and 2.20 GHz), which means it's a bug and will take some time to fix. I hope that's not the case.

---------- Post added 18th December 2011 at 02:31 AM ---------- Previous post was 17th December 2011 at 02:53 PM ----------

Well I finally fixed the problem in an extreme way: I built a custom kernel with cpu scaling disabled. That appeared to be the only way to fix it. The build did not go completely smoothly and I had to do a bit of manual tweaking to get the new kernel to boot. But now I get the maximum frequency:

$ uname -morv
3.1.5-2.custom.fc16.x86_64 #1 Sat Dec 17 20:19:46 EST 2011 x86_64 GNU/Linux
$ cat /proc/cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 15
model           : 55
model name      : AMD Athlon(tm) 64 Processor 3700+
stepping        : 2
cpu MHz         : 2199.488
cache size      : 1024 KB
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt lm 3dnowext 3dnow rep_good nopl pni lahf_lm
bogomips        : 4398.97
TLB size        : 1024 4K pages
clflush size    : 64
cache_alignment : 64
address sizes   : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp

I tested some cli apps and they all now run twice as fast. Yay!

[PabloTwo]

Glad to hear your cpu is back up to speed.

[Dangermouse]

I set mine to max all the time by turning it off in the bios (cool and quiet) massive speed boost.

[DBelton]

same way I had to do it here, Dangermouse.

But in my case, I am running a processor a little overclocked, and the ONLY way I could get it to run at the overclocked speed was to disable the EIST (Intel SpeedStep) in the BIOS. cpufreq still sees the speed the processor is rated at, but it runs at the overclocked speed.

Without disabling SpeedStep, cpufreq would scale the processor, and then would refuse to set it any higher than than the maximum the processor is rated at.

Whoever wrote cpufreq does not want anybody to overclock, makes me wonder if they work for intel or amd or are getting paid by them to prevent overclocking.

[RupertPupkin]

I was thinking of filing a bug report but it looks like this bug has affected other users and reports have been filed:
https://bugzilla.redhat.com/show_bug.cgi?id=772649
https://bugzilla.redhat.com/show_bug.cgi?id=715329

I hope the Fedora devs remove cpu scaling from the Fedora kernels until this bug is fixed, and make it an optional service as before. That would be the considerate and right thing to do, otherwise they'll be slowing down computers that need top performance from their cpus. Having to build a custom kernel is not the right solution.

[amturnip]

Rupert, your readings and experiments might make a useful addendum to the appropriate bugzilla entry.

[merilius]

I think an even more annoying "feature" of F16 and gnome3 is that CPU Frequency Scaling Monitor from gnome-applets is now unable to change governors... I used it in F14 without any problems.
Now it displays the current frequency and governor but choosing a new one has no effect. I am not sure if this is because of cpufrequtils being removed (obsoleted) by kernel-tools or something else.

Right now on my desktop machine I need to cat a governor to some /proc file. Very user-friendly...