Aaand there is the Kaveri2.0 codename again . It turns out it's a real thing after all huh .
Edit:
As for clockspees, I doubt there is a radical drop in actual base/turbo from Richland parts.
Just look at desktop parts 1st. Top Richland is 4.1Ghz/4.4Ghz part and Top(?) Kaveri will be 3.7Ghz/??Ghz. My guess for Kaveri 7850K's Turbo is that it will be around ~4Ghz. So ~10% lower base and 10% lower Turbo Vs RIchland. Performance will similarly be up to ~20% vs desktop Richland (~30% pure performance jump at the same clock adjusted for ~10% deficit in clock speed).
On Desktop, ST performance will be lower of course, my guess is ~10% better vs stock Richland (~4.8Ghz "Richland equivalent" performance Vs 4.4Ghz stock Richland) and MT wil be ~20% better( ~5.1Ghz "Richland equivalent" performance Vs 4.2Ghz stock Richland since RIchland Turbos on all cores too in MT benchmarks).
On Mobile, if we assume similar clock speed disparity, top mobile part will be ~2.3Ghz base and 3.1-3.2Ghz Turbo. iGPU will similarly clock at 18% lower level Vs top mobile Richland part, at ~600Mhz.
Finally for AMD to achieve these "up to" performance numbers with ~10% lower clocked chip, it needs to have "up to" 30% better performance in those same benchmarks at the same clockspeed . Coincides with the slides we saw last year (up to ~30% more ops/cycle: up to 20% ST IPC x 10% due to decoder BW doubling).