Test System

Initial Setup

I updated the board's BIOS to the most recent version before running tests with the AMD A10-5800K. The AMD A10-5800K ran with 1866 MHz AMD Radeon memory for testing. I used the Kingston HyperX Beast sticks with the AMD A10-6800K. Since the Kingston sticks support both 2400 MHz and 2133 MHz via XMP profiles, and the ASRock FM2A85X-ITX supports XMP profiles, the combination of all three made for a pretty flexible set that gave me no problems at all. I'm actually rather impressed by how well this small ASRock board ran everything, including overclocking.

Overclocking

Overclocking with the AMD A10-6800K was great fun. I easily hit 5.0 GHz on the CPU core with 1.525V. I also managed to clock the GPU core to slightly over 160 MHz from stock, for 1014 MHz, and bumped the memory up to its native 2400 MHz. AMD mentioned that bumping up the Northbridge speed while increasing the GPU speed could also help, so I bumped that up to 2400 MHz to match the memory speed. The Northbridge ran at 1.35 V as dictated by how the ASRock FM2A85X-ITX read the XMP profile of the Kingston DIMMs.

Power Consumption

We measure CPU power consumption since one of our first tasks is to truly verify system stability. I isolate the power coming through the 8-pin ATX connector using an in-line meter that provides voltage and current readings, and total wattage passed through it. While this may not prove to isolate the APU power draw in all instances, it does serve as a good general indicator of the total power consumed by the APU itself while eliminating any power drawn by other devices, such as the motherboard, RAM, drives, and coolers. Measurements are taken while running wPrime 32M because it provides a consistently high level of power draw. This does not account for OpenCL-accelerated apps that utilize both the CPU- and GPU portions of the APU.


During OpenCL workloads, I recorded a maximum power consumption of 147W. Given these figures, we can surmise that the GPU portion at 1014 MHz consumes 41 W. I saw an even 110W at stock, giving the GPU portion about 30W of power consumed at peak under OpenCL. This is a bit higher than the 100W TDP, but given the inconsistent loading pattern, it's nothing other than worth mentioning.