Shown above is the A8-3870, currently the highest specification chip in the A-Series of APUs. It shares the same heatspreader design as all recent AMD desktop processors and underneath we find the socket FM1 pin layout for A55/A75 based motherboards.
This is a 32nm, 100w TDP chip with four cores which run at 3.0GHz using a 30x multiplier at 100MHz. The chip has 4x64Kbytes of L1 data and Instruction cache with 4x1MB of L2 cache and elsewhere on the die we have a DX11 GPU in the form of the Radeon HD6550D which again is a 32nm part. This GPU based on the 5000 series of desktop products has 400 "Radeon Cores", 8ROPS, and a 64-bit memory bus. The GPU has access to 512MB of the systems DDR3 and at stock runs at 600MHz.
All of the standard Radeon features are present such as support for DirectCompute 5.0, AVIVO and HDMI 1.4/Blu-Ray 3D. Finally, also contained within the APU is a Northbridge which would previously have been separate to the CPU.
What makes this chip different from the existing models and worthy of the "K" branding? That is its ability to overclock. AMD have unlocked the multiplier on these APUs and allowed fine tuning of the GPU speed so we can now overclock without bus speed changes and target 3D performance enhancements specifically with boards such as the ASUS FM1 range, which we are using today, offering simple menu options in the BIOS to overclock from 600MHz up through 900MHz and beyond. AMD indicate that the GPUs will all hit around 960MHz (with 1866MHz memory) and multiplier overclocks will hit the 3500-3800MHz range. Our APU sample did both with minor voltage tweaks.