Fine Tuning Power Management
By default, you can’t change the power management features of a processor. There are no settings to tweak in the BIOS (unless you count the ability to turn power management on and off).
However, you can run applications like RMClock, CrystalCPUID, K10Stat, and PhenomMSRTweaker to customize your power management configuration. These applications allow you to override default settings used by Intel’s SpeedStep and AMD’s Cool'n'Quiet. The only requisite is making sure each of the power management options you tweak is actually enabled in your motherboard's BIOS.
What, exactly, can you accomplish with these applications? In addition to choosing the settings for each p-state, such as operating clock rates and voltages, you can also change transition time between p-states and the level of workload required for a p-state change. It’s easier to explain fine tuning p-states like this:
Changing upward p-state transition to a lower value means you switch from power-saving mode to performance mode faster, so there’s little delay when running applications. Changing downwards p-state transition to lower values means you switch from performance mode to power-saving mode faster, conserving power when you don’t need the performance any more.
Changing the upward workload level means the processor won’t switch to a lower p-state as long as the workload is below that threshold. Vice versa, changing the downwards workload level means the processor will only switch to a higher p-state if the workload drops to that level or goes lower.
How does transition time relate to workload? The lower the value you use for transition time, the faster you switch to another p-state once the processor hits the workload level you specify. Of course, if you use a higher value, p-state transitions occur at a slower rate.
For processors with more than one core, you can even specify how to calculate the workload. The options can range from average workload to the highest workload on one core to a minimum workload on one core.
The trick to striking the best balance between performance and power consumption is finding the right combination of p-states, transition times, and workload levels. The right combination will not only allow you to realize performance very close to what you might expect without power management enabled at all, but also effectively conserve power and reduce operating temperatures.
The Tools of the Trade
To measure the effects of power management on performance (and our fine tuning efforts), we need a handful of tools.
In addition to the usual benchmarks and software utilities mentioned above, we need a tool to log throttling. Why? To see whether or not unnecessary throttling is happening. Luckily, RMClock provides this feature. Unfortunately, it lacks the ability to independently manage the power management features of newer AMD processors. We also need a power meter in order to actually measure the power we save. A Watts Up? PRO power meter is used for this purpose.
In this article, we are going to use six AMD processors: the Athlon X2 7750 Black Edition, Athlon X2 7850 Black Edition, Athlon II X2 250, Phenom II X3 710, Phenom II X4 945, and the Phenom II X4 955 Black Edition. Because RMClock is unable to manage power management features of these processors, we have to use either K10Stat or PhenomMSRTweaker. We opted to use K10Stat because it has the ability to manage more than just one p-state.
There are some architectural and design differences between these processors, but we won't discuss them in detail here. The Athlon X2 7750 and 7850 are still built on AMD's 65nm process, while the others are manufactured on the new 45nm process. The Athlon X2 7750, 7850, and Athlon II X2 250 are true dual-core processors, with no disabled cores or cache. The Phenom II X3 710 is basically a “failed” Phenom II X4. AMD has disabled the fourth core (with its corresponding L2 cache), although the 6MB L3 cache remains intact.
Note: We've retested the Athlon II X2 250 and Phenom II X4 955 on a new platform, swapping in Biostar's TA790GX A3+, which is a dual power plane Socket AM3 motherboard, and four modules of Team Elite DDR3 memory. The power supply was also swapped to a lower wattage unit (Enermax's 405W Tomahawk). We also added an Athlon II X4 620 sample to our data, which we've compared to the Phenom II X4 955 BE.
Here's some additional info on these processors:
|Core Clock||Northbridge Clock||L2 Cache||L3 Cache|
|Athlon X2 7750||2.7 GHz||1.8 GHz||2 x 512KB||2MB|
|Athlon X2 7850||2.8 GHz||1.8 GHz||2 x 512KB||2MB|
|Athlon II X2 250||3 GHz||2 GHz||2 x 1MB||N/A|
|Phenom II X3 710||2.6 GHz||1.6 GHz||3 x 512KB||6MB|
|Phenom II X4 945||3 GHz||2 GHz||4 x 512KB||6MB|
|Phenom II X4 955||3.2 GHz||2 GHz||4 x 512KB||6MB|