cancel
Showing results for 
Search instead for 
Did you mean: 

[X670E Hero] System often fails to POST

PainGod85
Level 7

Hi there! I've been running into a very interesting number of issues on my system. Specs as follows:

AMD 7900X3D
G.Skill Trident ‎F5-6000J3238F16GX2-TZ5NR (on QVL)
RTX 4090

First of all, I built it sometime in July. I immediately updated the UEFI to v.1516 due to the issues cropping up with X3D CPUs exploding.

Still using Buildzoid's RAM timings, the system had no issues booting up on a daily basis until around two weeks ago. That's when the system would start telling me that a boot attempt had failed and safe presets had been loaded.

My next step was checking to see whether the system would boot with EXPO I enabled - which it did, for a few days. Then the same behavior started to crop up.
I then set the system to use EXPO II just to see whether the issues would go away - which they did, but again only for a few days.

Having failed to get the system to reliably POST on 1516, I flashed 1709, and much of the same is happening now. EXPO I was stable for a few boots, then EXPO II, now the same issue persists.

The system will usually boot normally until Q-Code 70, display 00 for a few fractions of a second, then reset and attempt to POST from scratch. Sometimes it gets stuck. Almost everytime afterwards, I end up with the message that a previous boot attempt had failed.

Under 1516, it got stuck at least once with code c5.
Earlier today (with 1709) it got stuck with 15, 1a and 0d.

I would like to stress here that the system is completely stable once I'm in Windows. I haven't had a BSoD ever, whatever I do with it. Additionally, safe presets work without issue.

This leads me to assume it is some sort of initialization issue, almost certainly related to RAM, probably has something to do with DDR5 memory training, and is not at all a stability issue.

Especially on 1709, I've tried to enable and disable both Memory Context Restore settings found in the two places of the UEFI (WHY?!), without success. I've also tried to enable DFE Read Training, also with limited success. I'm pretty sure I tried all variations of these under EXPO I and II.

At this point, I am however at a loss. I've run out of all options I could think of on my own,  so if anyone has any suggestions on what the issue may be, or if anyone has successfully troubleshot the same issue, please help me.

368 Views
3 REPLIES 3

Silent_Scone
Super Moderator

Hello,


@PainGod85 wrote:

Hi there! I've been running into a very interesting number of issues on my system. Specs as follows:

AMD 7900X3D
G.Skill Trident ‎F5-6000J3238F16GX2-TZ5NR (on QVL)
RTX 4090

related to RAM, probably has something to do with DDR5 memory training, and is not at all a stability issue.

 

Hi, 

Training has everything to do with stability. The best way to think of it is is that if the system *did* post rather than fail at the current attempt, it's likely the system would be unstable. Crudely it implies the overclock does not have sufficient signal margins to pass what is defined as electrically valid.

If it only started happening recently, it may be worth remounting the CPU and cooler (reapply TIM) to ensure it's not a pin pad contact issue, if not:

Firstly look at tuning VDDIO and VSOC and see if the situation improves.

 

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

I'll try adjusting VDDIO first. In another thread about similar issues with a different 600 series board, that seems to have helped the user. EXPO has that voltage set at 1.35V for some reason, and I've now reduced it to 1.25V.

Two reboots in, it's working, but of course the sample size is negligible so far.

Funny how one would traditionally associate higher voltage with better stability at the cost of heat generation and possibly component longevity, but with DDR5 and its associated controllers - at least on the AMD side - increasing it beyond a certain point may actually be counterproductive.

No worries, a lot of users fall prey to this misconception. More voltage can hurt just as much here due to signalling-related issues, that's why "tuning" is used and not necessarily increasing voltage. Some CPUs are simply more comfortable at a lower VDDIO voltage than others.

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090