To follow up on the in-meeting roundtable the other day:
I found some newer reports that ECC can work on Ryzen, if the stars are aligned.
http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/75030-ecc-memo...
Basically says they got to the stage where EDAC in linux recognized the ECC and could get single-bit errors generated and logged. That's all I'm interested in, really (double-bit errors don't concern me as much, esp if they're still logged). So this looks promising.
Looks like you have to be quite selective in mobo, and the one they tested was: ASRock X370 Taichi (not cheap!)
I wouldn't buy a Ryzen mobo for ECC support unless someone had already tested it like they did to confirm it supports ECC.
So it's nice to see it is possible, though the support & guarantees are pretty dodgy. Ah, back to the good (bad) ol' days (2000-2005) of VIA chips and Intel boards that maybe/maybe-not had ECC that actually did something.
===
If you want guaranteed (validated) ECC on Ryzen, looks like you go with the recently released Ryzen Threadripper 8c/16t CPUs and HEDT "high-end desktop" boards. However, then you are looking at prices like Intel "workstation" products, though you will get more threads/buck.
Personally, I don't have a use for tons of threads (yet), so for me a lower-end Xeon with a workstation board might on the table again, unless I want to take the risk with a maybe/might/should-work regular Ryzen.
Adam's article (coming soon!) on Ryzen random mem latency for routing might be the straw that breaks the Ryzen camel back for me. We'll see...
As always, if someone beats me to building (from parts) an ECC *workstation* (i.e. real, full x16 slot) in '17 or '18, post your results here!!