Originally Posted by nightfire

In case anyone isn't following the overclocking thread (pretty high traffic), if you're looking for even more performance (at the potential cost of some battery life, though I haven't observed any diference so far), try this (copied from my post there):

Code:

echo 75 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/up_threshold
echo 150000 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate

By default, the ondemand governor is very aggressive about when it cycles up (waits for 95% load), and can have a significant latency of up to 300ms. This means if you saturate your CPU, your average wait for frequency increase is 150ms (0-300ms).

This change cuts that time in half, and lowers the bar for how much load is needed before committing.

It may affect battery life slightly (more frequent sampling, more aggressive cycle-up), but gives a yet more snappy feel. It may have no effect (or even lower battery consumption) by minimizing unneeded time at high cycle rates (responding to drop in load more quickly). Try it out!

If you want to make it permanent, just do this:

Code:

cat > /etc/event.d/ondemand-config << EOF
start on started xsession

console output
script
        echo 75 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/up_threshold
        echo 150000 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate
end script
EOF

In case anyone is wondering "why on started xsession?" .. it's because at some point late in the bootup, the ondemand scheduler is selected by something and I'm not sure what. When it's set, it resets the values. So I just wait a bit.

Perfect complement to any overclocking kernel.

Originally Posted by nightfire

In case anyone isn't following the overclocking thread (pretty high traffic), if you're looking for even more performance (at the potential cost of some battery life, though I haven't observed any diference so far), try this (copied from my post there):

Code:

echo 75 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/up_threshold
echo 150000 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate

By default, the ondemand governor is very aggressive about when it cycles up (waits for 95% load), and can have a significant latency of up to 300ms. This means if you saturate your CPU, your average wait for frequency increase is 150ms (0-300ms).

This change cuts that time in half, and lowers the bar for how much load is needed before committing.

It may affect battery life slightly (more frequent sampling, more aggressive cycle-up), but gives a yet more snappy feel. It may have no effect (or even lower battery consumption) by minimizing unneeded time at high cycle rates (responding to drop in load more quickly). Try it out!

If you want to make it permanent, just do this:

Code:

cat > /etc/event.d/ondemand-config << EOF
start on started xsession

console output
script
        echo 75 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/up_threshold
        echo 150000 > /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate
end script
EOF

In case anyone is wondering "why on started xsession?" .. it's because at some point late in the bootup, the ondemand scheduler is selected by something and I'm not sure what. When it's set, it resets the values. So I just wait a bit.

Perfect complement to any overclocking kernel.