Continuing with the Llano overclocking discussion, we investigated how big of a performance gain overclocking of this APU may produce for us. To create a complete picture we compared the performance of AMD A8-3850 working in nominal mode at 2.9 GHz frequency against its performance in three different overclocked modes. The first one was a “straight-forward” mode when we simply increased the BCLK frequency but used the default x29 multiplier. The second mode was “intelligent” mode when the BCLK frequency was increased to its maximum (141 MHz in our case) and the processor clock multiplier was set at x26. And the last one was the “creative approach” mode when the focus was on reaching the maximum memory frequency (DDR3-2133 in our case), which required lowering the BCLK frequency to 133 MHz and increasing the processor multiplier to x27.
Moreover, we ran all tests twice for the configuration with 133 MHz base clock generator frequency: with the memory working at 2133 MHz and with the previous multiplier – as DDR3-1774. Comparing these results will allow us to conclude how efficient memory overclocking is in Socket FM1 systems irrespective of the effect from other components frequencies.
As a result, the diagrams below show 5 columns corresponding to the following operational modes:
To estimate the average system performance we used PCMark 7. It measures the speed of typical real-life algorithms that are very popular in every-day tasks.
The computational SuperPi test is a great way of checking the performance in single-threaded mode. This test calculates 32 million digits of the Pi:
Multi-threaded performance during intensive calculations was checked using wPrime 2.04 test.
To test the performance during data archiving we took a benchmark built into WinRAR 4.0 archiving utility.
Final rendering speed was tested in Cinebench 11.5.
The x264 HD Benchmark 4.0 on the diagram below transcodes a small video clip in two passes and the entire process is then repeated four times. We are providing average results of the second pass.
We measured the performance in Adobe Photoshop using our own benchmark made from Retouch Artists Photoshop Speed Test that has been creatively modified. It includes typical editing of four 10- megapixel images from a digital photo camera.
To check out the performance during HD-video transcoding we measured the time it took to transcode a 10-minute H.264 1080p video clip into an iPhone 4-friendly format in lower resolution. We used a popular commercial utility from Cyberlink called MediaEspresso 6.5, which utilizes computational resources of the graphics core via AMD Stream technology.
Judging by the results of typical processor tests we can conclude that Llano overclocking makes a lot of sense and you shouldn’t neglect it if the opportunity presents itself. 26% increase in the processor frequency from the nominal 2.9 GHz to 3.66 GHz allowed to boost the performance almost proportionally: by 25% on average. At the same time, when you overclock by raising the BCLK frequency and at the same time reducing the processor clock multiplier delivers a guaranteed better result, even though the memory doesn’t work at its maximum speed at 141 MHz BCLK. Overall, memory frequency doesn’t affect the purely processor tests that much at all. The performance difference will be on average 2.5% when you change the DDR3 frequency multiplier by 1 step.