Just sharing some experiences with my Ryzen 1700 build, as testing have somewhat settled down.
I'm presently satisfied with 3.8 GHz at an offset VCore of +0.150 and a fixed SOC at 1.20V. Which, importantly, results in a measured top VCore of 1.373 when stress testing using AIDA64. 3.9 GHz on this individual chip seems to require a real world voltage of well over 1.40 which doesn't make me comfortable right now.
I went for the offset VCore rather than fixed in order to achieve a lower power consumption for everyday use, which in turn keeps the noise at a minimum outside high loads. (it drops down to 0.53)
It's possible to run an offset for the SOC voltage as well, but I had stability problems when trying that for a bit and haven't found a way to measure SOC voltage live within Windows either. It's not a huge amount of resulting power either way.
I expected to but never used AMD's Ryzen Master overclocking software, mostly because my Windows installation complains about an unsigned driver in the software package. Otherwise I guess it could have shortened my testing period a bit. It hasn't helped that for whatever reason my gaming monitor won't display the BIOS picture right unless I switch from DisplayPort to HDMI, which is a bit...
On the RAM side I might have been a bit too eager to build me a system with 64GB of RAM, as I seem to be left with running my four sticks at 2133. But that's also what you can end up with being an early adopter of a platform, not waiting for reviews and RAM certifications to be finalised. Still loving the 64GB of RAM. And 2133 is still above the officially supported Ryzen RAM speed for four dual ranked DIMM's, so I can't knock this Asus Prime X370 Pro motherboard for it either.
On the sofware testing side I have been using Process Lasso a lot in order to follow the behavior of programs and games. It's been both informative and fun (has geek much?), but yesterday I decided it was okay to forgo using it. The thing is that even if the Windows thread scheduler does sometimes put a thread of for instance a game on the "logical" core instead of the physical one, I haven't seen any resulting practical performance loss on my system (it does not double-load both the physical and the logical). And we can't mitigate the performance limit of the Infinity Fabric by juggling cores anyway.
As I'm gaming at 1440p I don't think that I'll have any reason to revisit this factor, until I've gotten my dual Vega cards. (now I'm kidding...)