At the end of March I also published some basic Nouveau power consumption benchmarks, with this mini article today continuing in that theme.
While for a mobile NVIDIA Quadro FX880M with the Nouveau driver the power consumption is on-par with the proprietary NVIDIA Linux graphics driver that properly supports various power management features (e.g. NVIDIA PowerMizer), when disabling kernel mode-setting to fall-back to the VESA driver, the numbers aren't too different either. With the VESA driver though you only have the Gallium3D LLVMpipe driver for acceleration, which isn't really beneficial for end-users unless having a strikingly-fast multi-core processor. (Additional details.)
The only major open-source driver left still with some level of user-space mode-setting (UMS) support is the Radeon driver. Intel quickly killed off their UMS support soon as their kernel mode-setting support began to stabilize in 2010. The Nouveau UMS support was killed long ago before it was ever stabilized and the developers realizing KMS is the future for Linux graphics. With the Radeon driver there is still upstream user-space mode-setting support, but with recent hardware launches AMD has only been enabling new hardware support via the kernel mode-setting paths. So for around Evergreen GPUs and older the kernel can boot with nomodeset or radeon.modeset=0 and the xf86-video-ati driver can still provide mode-setting support.
While the Radeon X.Org driver can still handle user-space mode-setting for most hardware, the Gallium3D driver on the Radeon side is only supported along the DRI2/KMS paths. In other words, with Radeon UMS you'll be dropping back to the LLVMpipe driver for CPU-based 3D acceleration since the classic R300/R600 Mesa drivers that did support acceleration under UMS were since removed. So it's basically useless to be using Radeon user-space mode-setting, but as the graph above illustrates, in a stock configuration it doesn't provide any power-savings advantage now that the Radeon KMS support is mature. (The results on OpenBenchmarking.org.)
As the quick Radeon UMS/KMS testing was done from a notebook with a mobile Radeon R500 (X1000 series) graphics processor, no Catalyst driver comparison could be done from this ThinkPad since the proprietary driver no longer supports this aging ASIC.
For those curious about the power consumption between the open-source and proprietary drivers for AMD and NVIDIA graphics cards in fine detail, see The Most Comprehensive AMD Radeon Linux Graphics Comparison and A 14-Way Comparison Of NVIDIA vs. Nouveau Drivers. There's GPU power consumption numbers in there along with the performance results to look at efficiency. Those results are a couple months old, but not much in the way of power management has improved since then, although new benchmarks will be conducted in the near future thanks to performance gains.
Coming up in the next few days will be some revised power numbers from various desktop environments of Ubuntu 12.04 LTS, as hinted at yesterday.