Short answer is we don't know. This is something I've been attempting to follow for a while as far as AVX/FMA performance is concerned.
Compared to current AMD CPUs, we can at least expect about double performance per core, since there will be one FPU per core, not one per two cores as current. This is of course on the assumption they're no worse than the current FPU.
Double might sound good, but it isn't. Due to their implementation of instructions used for higher speed, per FPU current ones can't come close to Intel. This is the big unknown in their future FPU performance. Based on slides known to come from AMD, I'm leaning towards that being it. There remains an optimist's view they have done more than that, and it will actually catch up with Intel. We will have to wait and see.
In the best case, that they now have FPUs comparable to Intel's, there will be another different problem. Ryzen for consumer appears to only offer dual channel ram. If all the FPUs on an 8 core part are that fast, it will be trivially easy to be ram bandwidth limited for large LLR tasks. It would destroy small LLR tasks though...
The existing demos shown publicly by AMD don't give us anything useful to relate to FMA performance. If pricing and availability isn't silly, I intend to get one as soon as practical after launch to do my own testing. I'm sure there'll be mainstream reviews shortly before that, but they're unlikely to cover it in a way useful to us.