Mobile Users Beware: Linux Has Major Power Regression
As the Phoronix Test Suite was designed to be extremely extensible, the system monitoring module can also poll some other system vitals -- from the system I/O wait down to the CPU fan speeds and temperatures. Here is a look at some key facts between the Linux 2.6.37 and Linux 2.6.38 kernels when running the OpenArena game.
The CPU usage isn't vastly different between 2.6.37 and 2.6.38, so it's not that there's some user/kernel-space interaction problem resulting in a CPU load problem like before when another big regression was spotted (that time the regression was fixed before making its way into a released kernel, but this time it's already been around for one development cycle without being addressed).
Enhanced Intel SpeedStep Technology also is not broken. With the 2.6.38 kernel there is an increase in flipping between the two speed states on this Core Duo CPU, but it's not the entire time so it doesn't explain the increased power consumption for the entire duration or for the increased power usage when idling. A troubled EIST would also likely lead to different results in the tests themselves, which was not the case.
The GPU frequency was also unchanged between kernels. Again, I have reproduced this problem on mobile devices with Intel, NVIDIA, and ATI graphics adapters. The full monitoring results are available in this other OpenBenchmarking.org result file. There were also no changes in the display's screen brightness.