Llano APU (tweaked K10h+GPU):
32nm HKMG SOI process.
11 metal layers ,dual strain liners, eSiGe ,low K dielectric.
35 million transistors and 9.69mm^2 (core without L2).
110 million transistors and 17.7mm^2 (core with L2 and power gating ring).
Performance improvements:
1.Instruction window is enlarged to 84 entries.
2.Instruction scheduler enlarged to 30 entries for Integer.
3.Instruction set is cleaned up (added AVX support?).
4.L1 cache cell 8T design for low voltage and good scaling.
5.L2 cache up to 1MB, 16-way associativity.
6.Improved hardware integer divide.
7.Reduced latency for FP instructions.
8.Better prefetcher.
9.Faster cache lines transition between states.
10.Increased memory fill speed.
11.TLB improved for better residency.
Power improvements:
1.Core Power Gating.
-added power gating ring bulid on NFET transistors
-ability to completely disconnect any one of the cores
-multiple power planes
2.Digital APM Module.
-digital monitoring amperage and temperature
-turbo functionality
3.Clock grid
-depopulated and power aware clock grid design
-80% reduction in clock grid metal capacitance
-50% reduciion in the number of power buffers
-2x reduction in clock switching power