After analyzing some threads and asking some questions on overclock.net I
finally realized that many people with this board are running
much lower SoC values than me.
Playing around with good manual/low timings for my Samsung b-die sticks at 3200MHz I had problems getting it stable using SoC 1.11875, procODT 53.3, DRAM 1.38, LLC Auto, extreme phase control for cpu and SoC. I got sporadic errors during my hci memtests at various stages.
Anyway, after reading these threads on overclock.net for our board I lowered my SoC to 1.0 (some people are actually running lower than SoC 1.0 for 3200MHz memory speed).
What's interesting:
After changing to SoC 1.0 I no longer get any hci memtest errors and achieve good benchmarks results.
Warning: I haven't tested my system using prime95 or occt after running SoC 1.0 though but it for sure looks promising, hci memtest should be good enough for testing memory stability?
Bios 0902, cpu 1600X (SoC value is probably very cpu dependent)
Running high memory speeds with good manual timings provides good performance boosts for Ryzen systems. Only thing I have been focusing on lately is tweaking my manual timings. Problem is that XMP/DOCP sets terrible timings, manual timings is a must. Reason why XMP/DOCP sets terrible timings is probably because the issues with Hynix memories, trying to achive Hynix compability. Running these default loose xmp timings with Samsung b-dies is
very uneffecient from a performance point of view.
Timings recommended from @The Stilt (overclock.net) is a good start even though he has the Crosshair board:
https://www.pixeltopic.com/files/2017/10/mxsmriiygmeolyr.jpg
The Hynix - Samsung story for Ryzen is tragical, AMD could have handled this better but Samsung is dominating, we can all agree upon that.