As promised, now that Linux 3.11-rc1 has been released, it's time for the new dynamic power management support of the Linux 3.11 kernel for AMD Radeon graphics. This first article previews the possible OpenGL performance gains for an AMD APU when enabling "DPM" for allowing the graphics core to properly re-clock based upon its workload.
Setting up the Radeon dynamic power management support is simply a matter of getting a Linux 3.11+ kernel (for Ubuntu users, there is the Ubuntu mainline kernel PPA with daily packages as well as the release candidates), fetching the latest AMD microcode files, and then passing "radeon.dpm=1" as a kernel module parameter. Linux 3.11 by default isn't enabling the Radeon dynamic power management so sufficient testing can occur and it will then be enabled by default in a later update. Via the kernel dmesg and sysfs/debugfs entries you can verify whether the Radeon DPM functionality is in use.
Radeon dynamic power management is a big win for mobile/laptop users where battery life is important and for higher-end Radeon GPUs that can be quite power hungry. Now when the GPU isn't enduring high load, the GPU core and memory frequencies can drop to lower power states that also reduce the voltage, while then clocking back higher when encountering any 3D/OpenGL workloads. The dynamic power management support is also important since for AMD APUs and higher-end modern GPUs the default boot clock speeds can be lower than their factory-rated frequencies, which up until now weren't forced by the Radeon DRM by default when encountering 3D load. Now Radeon DRM users can maximize the full potential of their graphics hardware.
In this article are some Linux 3.11-rc1 benchmarks of an AMD A10-6800K "Richland" APU with Radeon HD 8670D graphics. The Linux 3.11 testing was done with and without the "radeon.dpm=1" power management support. Testing also happened from a vanilla Linux 3.10 kernel with its stock settings. Mesa 9.2-devel was in use with the R600 Gallium3D driver atop this modified Ubuntu 13.10 installation. All benchmarking was handled in a fully automated and streamlined parameter using the open-source Phoronix Test Suite software.