Originally Posted by skullbringer
feedback, comparing 1002 with 0083, since I wanted to go for max ram oc:
- pstate oc with non stock ref clock works good, managed to get 123.2 MHz x 32.5 just like 100 MHz x 40.0 with 1.4V (all with llc auto)
- pstate 0 multiplier cant go below 32 with fid 80, so for example 4 GHz with 135 MHz refclock is not possible atm
- offset voltage with pstate oc also worked as expected, now running + 0.325V for 1.4V under load
- ram oc is now stable @ 3600MHz 14-14-14-34 1.5V, where with the old microcode it would only pass memory training every third time and then not really be stable (so I had to run 3200 14-34 with 1002)
- ram training completes with 14-14-14-34 timings and 1.5V at speeds up to 3775MHz
- !!! ram oc over 3650 MHz is not stable at all in windows and causes all kinds of memory errors! I'm getting blue screens with "memory_management", "page_fault_on_nonpage_memory", registry corruption, windows lock screen image changing colors, all kinds of the weirdest data corruption.
what I tried to mitigate this behavior:
- increase ram voltage up to 1.7V
- increase soc voltage up to 1.2V
- increase core voltage up to 1.5V
- set boot voltage up to 1.7V
- disable bankgroupswap
- setting 3700MHz with various memory multipliers and refclocks
- losen ram timings up to 20-20-20-40
- all combinations of the above
tldr 3650 MHz ram clock atm marks the tipping point from awesome performance to total madness and data corruption in windows.
If you have important data on your system, do not use these "testing" bioses and stay on the official version. Dont be mad, if your Windows install just craps the bed.
oh and here is a screenshot of my 24/7 with non-stock ref clock pstate oc, offset vcore and 3600MHz ram (no performance bias):
Warning: Spoiler! (Click to show)