Tweaking And Undervolting

Clearly, the default voltage settings are quite high and can use a good tweaking. But why stop there? Why not select the lowest voltage required for each p-state clock rate?

As a matter of fact, we did just that. However, we must remind you that each processor is different. You may be able to use the same settings as we did. But then again, you may not. You simply have to go through a process of trial and error to see what settings work for you while keeping things stable.

Note: originally, we didn't go into detail how these values were chosen. We decided on the following settings after testing the platform rigorously, running hours of several benchmarks, such as 3DMark Vantage, 3DMark 2001, SuperPi 4M, 8M, 32M, Prime95, WinRAR and a complete Cinebench R10 test run, without a hitch. At 800 MHz, running all Cinebench R10 tests took about 30 minutes, SuperPi 32M took roughly 1 hour 20 minutes, and the additional 3D rendering/video encoding took between four and eight hours.

Once those tests were completed, we retested the same values on a different motherboard. As a result, we unsurprisingly found that idle p-state values varied from board to board. For instance, originally we were able to run our Phenom II X4 955 processor at 800 MHz with just 0.6V core voltage on the Gigabyte 790GP-DS4H and -UD4. On the new platform, with a normal HT voltage of 1.2V, we could only get away with 0.75V.

Here are the custom voltage-optimized settings for each processor we tested.

Athlon X2 7750

Athlon X2 7850

Athlon II X2 250

Phenom II X3 710

Phenom II X4 945

Phenom II X4 955

We ran a barrage of stability tests for each of the above p-states (SuperPi, Prime95, Cinebench, 3DMark 2001), in addition to using each machine for several days. As far as we can tell, these settings are very stable with the samples on-hand. Of course, your experience may vary. Try using a small voltage bump (0.025V) to find the settings which work best for each p-state.

As you can see, we only change the core and northbridge voltage values. If you want more customized settings, you can also change the multiplier for each p-state.

Update: Experiences with Socket AM3 and Athlon X4 620

Below you can see the p-state values we were able to achieve on an AM3 platform with K10Stat. We've limited testing on this platform to just the Athlon II X2 250, Athlon II X4 620, and the Phenom II X4 955 Black Edition.

As noted earlier, we were unable to change the northbridge voltage values as we did with the AM2 board. There's still ample room for the core voltage alteration, though.

You would probably like to know the voltage values for each multiplier, right? Here's the voltage ramp data we gathered.

Athlon X2 7750

Athlon X2 7850

Athlon II X2 250