A Detailed Look At The ATI Linux Power Management

Written by Michael Larabel in Display Drivers on 11 May 2010 at 10:56 AM EDT. Page 2 of 4. 31 Comments.

Our first test was the battery-power-usage test profile within the Phoronix Test Suite that monitors the notebook's battery power consumption as the system idles at the GNOME desktop for a brief amount of time, then continues idling while the display is signaled off via DPMS, and then the display restored and MPlayer is automatically fired up to playback a brief video clip.

To no surprise, the default power profile did the worst and the dynamic power management option did the best. The ThinkPad T60 notebook was consuming about 17.7 Watts on average during this test where as using the DynPM option dropped this average to 15.9 Watts -- a savings of 1.8 Watts. Forcing the GPU to always run in the lowest state led to nearly the same power consumption at 15.9 Watts while forcing the high power state had an average of 16.1 Watts. Why the "high" power profile actually has a power savings compared to the default method without power management is explained in an upcoming graph.

Here is a look at the CPU temperature during the battery-power-usage testing, but it is not too interesting -- the numbers during the gaming tests carry more value.

This graph shows the core frequency of the Mobility Radeon X1400 during the battery-power-usage test. The high and default power management methods were the only times that the GPU ran at its rated 392MHz clock speed. However, why the "high" power management method is actually a power savings compared to the "default" method is shown here. With the high power management, the clock speed is the same until the display was signaled off via DPMS, which at that point it clocked down to the lowest 128MHz stepping. When the display is off, as long as a power management method is enabled for the ATI Radeon GPU, it will be bumped down to the lowest possible frequency.

Using the dynamic power management caused the core to operate at its lowest frequency during the idle process and then to modestly boost the core frequency during the video playback process.

For this article, we added support to the Phoronix Test Suite for monitoring the fence count on Radeon graphics processors and how many are being processed per second when being polled to reflect the GPU's activity.


Related Articles