This first set of results is again looking at the vanilla Linux kernels (2.6.36, 2.6.37, 2.6.38, 3.0, 3.0.1, and 3.1 Git) plus when applying the "pcie_aspm=force" option on the current near-final Linux 3.1-rc9+ kernel. The system was from a Lenovo ThinkPad W510 with an Intel Core i7 Q 720 with 4GB of RAM, 160GB Intel SSD, and NVIDIA Quadro FX 880M graphics.
This Intel Core i7 CPU does not support the energy performance bias (EPB) feature and its BIOS also doesn't expose ASPM support, but does work fine and does require it being forced in post-2.6.37 kernels. Now for these results from the prominent mainline Linux kernel releases. This first test is of the battery-power-usage test profile that looks at the power consumption while idling, then while idling with the display turned off, and then the display re-enabled and a brief video clip is played through MPlayer.
Even with the vanilla Linux 3.0.1 and 3.1 Git kernels, the power usage is still significantly higher than in pre-2.6.38 releases. From Linux 2.6.37 to 3.1 Git, the power consumption for the ThinkPad is up by 30%. Only when forcibly enabling ASPM does the power consumption on the modern kernel releases drop nearly to a level where it was in 2.6.36 and 2.6.37. When moving from 3.0 to 3.0.1 or 3.1 Git, which has the "x86, intel, power: Initialize MSR_IA32_ENERGY_PERF_BIAS" commit, there was no real change in power consumption since this hardware doesn't support the MSR in question.
The OpenArena story is much the same as the battery-power-usage results.
At least the Nouveau DRM driver is faster on the newer kernels, but that's already been looked at in numerous other Phoronix articles.