On Thursday 10 November 2005 18:59, Jonathan Brooks wrote:
Sorry for my ignorance of this topic, but does this mean that this bug in cpufreq will give much lower performance on these machines, since it will effectively nobble both cores making them run at 1GHz instead of 2.2GHz? Or will it just scale up according to need.
If only one core is loaded, both cores will toggle between 1GHz and 2.xGHz. If both cores are loaded, both will run at 2.x GHz (maximum frequency). To give an example: A benchmark that takes 1.3s with CPU frequency set to max, and 2.6s with CPU frequency set to min takes somewhere between 1.3s and 2.6s when set to "dynamic" (10 consecutive runs of the same benchmark): 1.55user 0.00system 0:01.57elapsed 99%CPU 1.31user 0.00system 0:01.32elapsed 99%CPU 1.34user 0.00system 0:01.36elapsed 99%CPU 1.80user 0.00system 0:01.82elapsed 99%CPU 2.66user 0.00system 0:02.68elapsed 99%CPU 2.64user 0.01system 0:02.66elapsed 99%CPU 2.63user 0.00system 0:02.63elapsed 100%CPU 2.63user 0.01system 0:02.65elapsed 99%CPU 2.65user 0.00system 0:02.66elapsed 99%CPU 1.40user 0.00system 0:01.42elapsed 99%CPU It *should* always take 1.3s, with the exception of the first run, which can take a little longer (powersaved needs some time to respond). -- Bernd Paysan "If you want it done right, you have to do it yourself" http://www.jwdt.com/~paysan/