There’s a stream tonight (UK time) where AMD will officially talk about Zen. Or if you can’t wait until then someone leaked the slides: http://videocardz.com/64741/ryzen
Ryzen Q1 2017
8 core, 16 thread
3.4 GHz+ with XFR (looks like more aggressive turbo boost like GPUs do to maximise work for cooling conditions)
20MB L2+L3 cache (512kB L2 + 2MB L3 per core?, recognising it may be different type than Intel)
Take away point from the video, for those that haven’t seen it and can’t be bothered, is that Ryzen fixed at 3.4 GHz was equal to an i7-6900k at stock in the AMD supplied Blender render file. They also showed a Handbrake test which showed Ryzen a bit faster.
They released the Blender file to allow anyone to run it, but with a mistake in one setting so times didn’t match up. Still, I tried to characterise Blender itself to see what performance it represents. Doesn’t seem sensitive to ram performance. Big bonus from HT on order of 50%. Doesn’t seem to use AVX. Each step from Haswell to Broadwell, then to Skylake, gains about 3% IPC. So, not really a heavy load like Prime95. I intend to test relative power usage of Blender on the weekend.