Red Hat Bugzilla – Bug 447975
2.5GHz Q9300 CPU frequency defaults to 2.0
Last modified: 2013-03-05 22:55:56 EST
Description of problem: Intel Q9300 quad-core cpu and DG33FBC motherboard + Fedora 9 64bit. CPU runs at 2.0ghz (add-to-panel CPU frequency scaling monitor for each core). Every time I boot the system, I have to change each core from 2.0Ghz to 2.5Ghz. Checked with BOINC and result was about 5500MIPS and when changed to higher cpu frequency result was about 6900MIPS. Version-Release number of selected component (if applicable): How reproducible: Every time Fedora is booted, CPU is scaled to 2.0Ghz as default. Steps to Reproduce: 1.Boot fedora9 64bit on DG33BFC with Q9300 CPU 2.Log on and check cpu frequency monitor for each core. 3.CPU has been scaled to 2.0Ghz but can be changed to 2.5 from the software. Settings can't be saved so after next boot same problem. Actual results: Q9300 CPU scaled to 2.0Ghz Expected results: Q9300 CPU frequency at 2.5Ghz Additional info:
ah, I bet this is one of those 'run loads in the background when idle' kind of apps.. For these, we have a tunable. try.. echo 0 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/ignore_nice_load echo 0 > /sys/devices/system/cpu/cpu1/cpufreq/ondemand/ignore_nice_load echo 0 > /sys/devices/system/cpu/cpu2/cpufreq/ondemand/ignore_nice_load echo 0 > /sys/devices/system/cpu/cpu3/cpufreq/ondemand/ignore_nice_load
Thanks, I just needed to enter the first line (echo 0 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/ignore_nice_load) and it changed performance for all cores (cpu1...etc don't even have the ignore_nice_load file). I tried putting this command on my /etc/rc.d/rc.local but it didn't work. Where should I put this command to run it after boot? In my humble opinion this should be the default and a lot of people - or should we say many BOINC projects - will end up with -25% CPU power. I didn't even notice significant power consumption difference (but I acknowledge that my wattage meter may not be completely accurate in measuring computer PSU's) so I don't see any point why the default should not be 100%.
actually, thinking about it, that _should_ be the default. Checking confirms, that is the kernel default. It looks like /etc/sysconfig/cpuspeed should control it, but even when IGNORE_NICE there is set to 0, something is changing it to a 1
I talked about this on the SETI@home website http://setiathome.berkeley.edu/forum_thread.php?id=47203 and they pointed out that some BIOSes now have the EIST. I disabled the "Enhanced Intel SpeedStep" from the BIOS and everything works.
So I tracked this down. This gets set to '1' when we log into gnome, when gnome-power-manager starts up. This code... on_ac = gpm_ac_adapter_is_present (cpufreq->priv->ac_adapter); if (on_ac == TRUE) { gpm_conf_get_bool (cpufreq->priv->conf, GPM_CONF_CPUFREQ_USE_NICE, &cpufreq_consider_nice); gpm_conf_get_string (cpufreq->priv->conf, GPM_CONF_CPUFREQ_POLICY_AC, &cpufreq_policy); gpm_conf_get_uint (cpufreq->priv->conf, GPM_CONF_CPUFREQ_PERFORMANCE_AC, &cpufreq_performance); } else { gpm_conf_get_bool (cpufreq->priv->conf, GPM_CONF_CPUFREQ_USE_NICE, &cpufreq_consider_nice); gpm_conf_get_string (cpufreq->priv->conf, GPM_CONF_CPUFREQ_POLICY_BATT, &cpufreq_policy); gpm_conf_get_uint (cpufreq->priv->conf, GPM_CONF_CPUFREQ_PERFORMANCE_BATT, &cpufreq_performance); } seems to want to only adjust settings when we're on battery, but as you can see, we do exactly the same thing for the nice setting regardless of power source.
*** Bug 453460 has been marked as a duplicate of this bug. ***
All this code got ripped out long ago from g-p-m.