5,000 rumors flying around...
There's only one metric that is semi-official, GlobalFoundries' +10% performance from 12nm LP vs 14nm on the same power , along with 15% area reduction (transistor density). It will also likely improve on Precision Boost 2 from Ryzen mobile, making XFR more potent when multi-threading & cross CCX latency.
Let's not overhype it. I'd rather be pleasantly surprised than woefully disappointed.
Right now Zen is around Broadwell IPC but lacking some clockspeed and AVX2 (AVX-256 bit is split into 128-bit) runs slower. Keep in mind a normal i7-8700k runs AVX at 4.3GHz (4.1GHz if with -2 offset) all core boost, with MCE and -2 AVX offset you're looking at 4.5GHz.
Due to Amdahl's law, I believe AMD's biggest issue with Ryzen is it doesn't clock very high so unless IPC is improved along with clocks
it needs a hefty clockspeed boost so that it isn't held up by a main thread. This will be slightly alleviated with an improved Precision Boost 2 & XFR. Unless power target bumps up to 125W from 95W then any larger gains (>10%) will need to be from tweaks to the Ryzen design.
meaning MATLAB would benefit from clockspeeds or IPC gain regardless of the AVX2 differences to Intel CPUs
& the poor threading of anything Autodesk including simulation:
& Euler3d CFD benchmark (memory reliant)
Keep an eye on Adobe (Premiere, After effects, Lightroom, Photoshop, etc), ANSYS, Autodesk (Autocad, Maya, Inventor, Revit, etc), and Dassault Systèmes (Solidworks, CATIA) results.