Measuring Stars core improvements
Anan's desktop Llano preview analyzed an A8-3850 APU. As this APU don't support any Turbo core states, it is the perfect candidate to measure the Stars core improvements as no other variables are at play here.
So, let's see the performance advantage that an A8-3850 APU (four cores @ 2.9 GHz) has over an AthlonII X4 635 CPU (four cores @ 2.9 GHz) in a single-thread Cinebench R10 test:
The benchmark was normalized to scores per GHz. As you can see, the microarchitectural improvements seems to give a ~8 advantage to AMD's new APU.
What about a multi-threads test?
We have a similar picture, with a ~7,5% better score for Llano APU.
Obviously, this >7% advantage is a better case scenario. Other test (Sysbench, x264, 7-Zip, etc.) show much smaller gain, with the Llano APU only between ~0-3% faster than the X4 635. However, is worth noting that many of the others benchmarks are to some extent influenced by disk/memory subsystem speed also, so they are not measuring CPU speed only.
So we can state that AMD published ~6% microarchitectural gains seems to be correct. In particular, take note of the ~7,5% Cinebench gain as we will shortly use this value to estimate Turbo core's operation in the A8-3500M mobile Llano chip.