Donanimhaber posted some desktop Trinity benchmarks a few weeks ago, the A10-5800K graphics were about 51% faster than the A8-3850's, which is inline with the 50% faster claim AMD has made. No idea if the benches are legit, though, take them with a grain of salt.
http://www.donanimhaber.com/islemci...on-A105800K-Trinity-islemcisi-test-edildi.htm
The only CPU benchmark than they ran was SuperPi, and it was only about 9% faster. SuperPi is pretty floating point heavy, though, other workloads may see a bigger benefit (AMD has claimed 20-30% better CPU performance I think, I'd assume this is "up to" and not seen in most workloads). And as speculated most of the benefits seem to come from increased clock speed, A10-5800K is supposed to have a 3.8GHz base clock and 4.2GHz turbo clock, for example, compared to only 2.9GHz for the A8-3850. But like pelov mentioned I don't think it matters much if they have to clock them high just to beat Llano in CPU performance, as long as they can hit these clocks while maintaining low power consumption, which it seems they can. It's sounding like improvements to GloFo's 32nm process and new technologies like the resonant clock mesh have really allowed them to increase the perf/watt of Piledriver cores, which is something they desperately needed to remain competitive.