AMD A10-6800K APU Review – Richland comes to Desktop

Introduction:

It’s been less than a year since AMD launched their desktop Trinity APUs and now they are launching Richland desktop APUs. Last year, Trinity brought some considerable improvements both in terms of performance and architecture as compared to Llano but the Richland is simply a more refined version of Trinity. Both have same architecture, same die-size, and same 32nm manufacturing process but Richland has more aggressive transistor tuning.

The improvements are mostly made on the software level. Power management is where AMD did most of the work. Both Trinity and Richland chips manage power states transition with the help of a 32-bit microcontroller which makes most of its major decisions based on the power consumption of the chip under any given task. On the other hand, Richland makes use of the several temperature sensors present on the chip and makes the clock speed changed depending the temperatures. Trinity also has these on-die sensors but it wasn’t really designed to take full advantage of this. Intel is also using similar temperature-based turbo scheme since Sandy Bridge.

Microprocessors usually take some time to heat up even if they are under full load so temperature-based turbo scheme helps in maintaining the low speeds for longer period of time hence saving the power considerably.

AMD Richland Die Shot

AMD has also improved that how power is distributed between the CPU and GPU parts of the APU. In Trinity, most of the power was routed to a single component even if it wasn’t really needed. However, the Richland is designed to intelligently route the power depending on the fact that if that particular part really needs that power. For example, if one part request for extra power for its turbo to kick in, the turbo controller will determine if that part is really bottlenecking the performance of other part or not. This helps in providing more balanced performance while keeping the power consumption to lowest possible levels.

These aggressive power saving schemes are of great help in mobile systems. On the desktop side, Richland doesn’t bring any considerable improvements apart from marginal performance increments which are mainly due to higher clocks.