X870E Nova D4 on restart/cold boot

Took the day to go properly over the situation.

Single dimm - 4d can occur (rarer then 2 slots)

GPU out - no change.

Took all nvme drives excpet windows system, same story.

Did alot and alot of manual running.

Patched 4.1 bios. Did alot ofanual adjustments to Processor droce strengths and impedances - and this greatly reduced 4d occurence.

By the end of the day, I spotted 4.2 update. Patched, so far tests go fine... We'll see, too early to tell.

What I can confidently say: error occurs only if memory training happened (code 15).

Will keep posted. For now slowly gathering info.

=======================================================================
BIOS v4.2 so far looks promising:

Did many restarts, restart from UEFI, online power cycle*, and couple offline power cycles**.

BIOS settings all bare-bone default, except: disabled Memory Context Restore and Fast Boot, so it is forced thru "memory training" (code 15).

So far so good, will keep you posted.

=======================================================================
________________
* - "online power cycle" - shutdown PC, power back on.
** - "offline power cycle" - shutdown PC, flip switch on PSU to "0" (so no more even 5VSB to mainboard) - power back on, few times after short time, once after long overnight period.

=======================================================================

Below goes what and how i tested before with older BIOS verisiony, v3.5 and v4.1 which i patched back after figuring v3.5 is not gonna solve the problem.

=======================================================================

BIOS v3.5:

While it seeemed to make a difference, most certainly it wasn't de-facto solution as 4d still appearead.

=======================================================================
BIOS v4.1
For my case, it proved to be quite terrible - but with manual adjustments to ODT, DQ and CA Impedances / Drive strenghts - it was much more reliable.

With manual adjustments, it would fail mostly when memory training was triggered.

Adjustmens (in AMD Overclocking submenu and AMD CBS submenu, NOT in OC-Tweaker section):
- Processor CA Drive streingths: 30 ohm
- Processor ODT Impedance Pull Up P0...P3: 40 ohm
- Processor ODT Impedance Pull Down P0...P3: 40 ohm
- Processor DQ Drive streingths Pull Up P0...P3: 34.3 ohm
- Processor DQ Drive streingths Pull Up P0...P3: 34.3 ohm

- All other BIOS settings default - with those v4.1 would throw 4d ONLY when it would do "long code 15" (memory training) (and then it wouldn't throw it always).
=======================================================================

Will keep you posted.

One more note. Whenver system was up and running, i would do alot of stress testing.
One example:
y-cruncher, zen5 stress test - vt3. Memory occupied by test: >59GB (while using IGPU, of the availble 61.something GB) - it would run for hours..

I also did swap dimm-slots, used single dimm, interchanged dimms. Even with single dimm 4d would occur (v4.1 and v3.5) - but less often.

Main GPU is still out, only one NVME (system drive, in PCIE-5.0 M2 slot) had been left in.

=======================================================================
PS. why is there no "edit" button on forum, previous post by me is a total late-night-written-hafl-asleep mess :)

PS2. When i say "all settings default", it truly means theat, except W_PUMP_FAN and AIO_Fan set to "silent", memory is NOT using any XMP / EXPO, aka memory is @4800MHz.

Well eventually it did happen.

PC was on for a bit, just idling.
Powered it off, next powering up - 4d.
Tried reset button - nope, 4d again.
Tried shutoff and cutting power via PSU switch -> 4d again.

Only thing that would bring it back was CMOS reset.

One thing that reliably worked EVERY time regardless of BIOS version was restarting via "ASRock Restart to UEFI" utility. That NEVER brought 4d - not with (v3.4, v3.5, v4.1, v4.2).

It's almost like if something somewhere in "training data" gets corrupted or some external influence, like small change in ambient temperature perhaps makes something drift slightly and it cannot get over with "training", after which CMOS reset brings it all back in place.

So I'm at a fence now, i like the board features, but not sure where to point finger at, certainly different BIOS versions behave differently. I'm undecided on another board - many lack the feature set this one has..

Stress tests it runs and runs and runs for hours, be it CPU / RAM / GPU or all-at-once test, so can it be a hardware issue?
It just looks like unfortunate (un)compatibility, or some bug present in "training" subroutines.

Would a following kit work on this board, is it known:
Crucial Pro DDR5 RAM 64GB Kit (2x32GB) 6400MHz CL40, XMP 3.0 & AMD Expo Compatible.

Part#: CP2K32G64C40U5B

My Kingston is also on QVL list, but only shown as "32GB", but even single dimm module throws 4d occasionally (less often so).

All good @Xaltar, you've been very helpful and surely i've been messy poster firing in million directions with my attempts, sometimes probably hard to understand what course of action am i taking.

I understand this single system is a way too-small sample to draw any conclusions - but if you'd ask me it does look like something in it throws whatever subroutines in training off. Be it "faulty" CPU / RAM / MBO.

Like i said, i like feature-set this board offers hence would like to go with it - the idea of going with different board is least favourable to me.

Pictures we can attach via URL, bb code?

=====================================================================
I'll post what is currently happening.

So after CMOS reset, as expected - system re-trained, booted into windows, default settings.

I used "ASRock restart to UEFI".

Changed two fan speeds to Silent.

This time, for the giggle, i used XMP1 profile. Just straight up XMP1, nothing else manually adjusted.

Noticed the board autmatically fixes VSoC to 1.2V whilst doing so and also OC-Tweaker Proessor ODT Impedance to 53 Ohms.

If you can get that answer, at which point during POST OC-Tweaker ODT impedance gets written into CPU's "memory controller"?

Earlier i asked about pictures - wish to attach one - system is running y-cruncher Zen5 stress test (XMP-1, which is 5600MHz for this kit), whole array of tests not just VT3, so both CPU and RAM are being stress tested.

It's been running for 40 minutes now, CPU avg temp 71°C, RAM avg DIMM_A avg 38°C, DIMM_B 38°C. Occupied memory during test: 61.7 GB..
Now would anyone consider anything "faulty" in there...

If i switch board, and it has slightly different memory training implementation, in case it's CPU that has issue it would be forever hidden..

Anyway, thank you for bearing with me once more. I'll keep my eye on beta bios versions in future, to test things out.


===============================================================================
PS. Not sure i mentioned earlier, ASRock support suggested trying out "LN2" mode enabled - it didn't really solve issue. I might re-try it again.