AGESA 1.0.0.6

First times I booted at 3066 cl16 or even 3200 I was tasting the victory.

I was wrong.

2.34 noticeably had problems syncing dram timings in OC page with Dram timing configuration in AMD submenu in advanced. Manual timings couldn't even being applied correctly, it was basically trash, but let forsee some potential.

2.36 better, no sync problems. But his behaviour is "not scientific". I cannot reproduce same results everytime.

I've pleayed with almost every new settings, including CLDO_VDDP, Geardown and powerdown modes, command rates and ProcODT . Simply not getting a clear scheme.

My best case with my sticks was 3200 CL 18-18-18-38 with timings borrowed from Asus Crosshair VI timings with the help of some guys with same stick as mine who was running 3200 CL16 flawlessy on his Crosshair agesa 1.0.0.6 beta (enough said..) but that with SoC 1.2v, dram at 1.43 (1.5v in windows) so waaaay more than what was needed IMHO. 
ProcODT working better were 48ohms and 53.3 ohms but never had full stability (problems sometimes in memtest DOS, sometimes in HCI, sometimes in GSAT, sometimes powerdowns).
I have experienced too a lot of new error codes, wich found to be PCH and Memory related. Also some 00 codes at stock frequencies or OC that I know my CPU is capable of.

I have even tried to input manually every (I mean Every single!) timing totally avoiding XMP reading.

The most important facts that made me SAD :
Not only I gained pratically nothing in AIDA and Maxmemm test going from 2933 to 3200 but most importantly GENERAL PERFORMANCES OF 1.0.0.6 BETAS ARE A COMPLETE MESS. 

Take for example BATTLEFIELD 1, with 2.34-2.36 @ default 2933 there is lag everywhere. It's like data are not properly transmitted among Cpu-Ram-PCIEX (same conditions, gpu utilization down from 85-90% to 72% almost fixed WTF? ) something not right.
And it is clearly visible in Firestrike Physics score too. Not able to get over 55 fps on the physics only.

Regarding SPD, I was beginning to think about buying Thaiphoon to do the same you thought. Writing proper manual timings on a separate profile.
The point is that I don't have an Intel platform to actually verify what the sticks are capable of to be sure the custom-xmp would work properly on the best case scenario, but most importantly, based on what I've seen in these betas, It would have been simply not worth the hassle.

I'm back to stable 2.34 STRONGLY HOPING that asrock engineers have some hynix M chips and got the same feedback as mine.

If any of you here are reading this, please report to asrock engineers . Something is not right in communication between CCXs, RAM and PCIEX .  I don't know what. But I am losing performance at the exact same settings as 2.30 agesa 1.0.0.4 bios.