Memtest86 9.4 troubleshooting in Z690P chipset (DDR5)

[eortizs]

Hi all,

I've have a situation with memory testing in DDR5, I am testing in same exact board model (ASUS PRIME Z690-P) in each situation and come across with different results on Memtest.

I have four test machines in which I have taken Test Machine 1 and Test Machine 4 as sample to experiment, I'm trying to test RAM stability on the following specifications 4800MT/s 36-38-38-83 1.25v VDD/VDDQ CR 2T, tweaking VCCSA and IMCV, and the results are fail in both cases, the odd thing about this is that I have Test Machine 3 which always passes this test spec. with the modules that fail on Test Machine 1 and Test Machine 4. Below is a link to wetransfer which includes the MemTest reports for Test Machines 1 & 4. I saw that theres a certain bit which always fails, also the reports on Test Machine 4 seem a little bit odd highlighting test 3 when the test that caused the abort situation was test 4. I am using latest BIOS available on the website.

https://we.tl/t-jphvRJcFMs

Other thing to note is that the errors concentrate in new Intel's E-cores and not in any of the P-cores.

Here's all the technical info. of the test machines:

Motherboard: ASUS PRIME Z690-P
BIOS: 1401
CPU: Intel i5-12600K
RAM: Micron IC DDR5

[David (PassMark)]

I can only assume the whole setup is marginal.
Meaning it is right on the edge of working / not working. So tiny differences in timing and voltage between machines are pushing it one way or the other.

[eortizs]

Hey David, thanks for your answer! While testing I've tried to mimic the same specs. of VCCSA/IMCV and even other DRAM related settings, so when testing every machine has same values on every setting.

Just as a follow up, as I know this Intel E-core technology is fairly new.

I have tried disabling the E-cores in Memtest using CPULIST parameter (leaving CPUs 12,13,14,15 out of the test) and so far 16 modules have passed the test without any errors.

Do you think this may be related to an enhancement needed for the software to be in sync with the new E-cores behavior? Or the E-cores failing in test mean that the modules are bad?

[David (PassMark)]

We don't think it is a issue in MemTest86. Nothing special should be required from application level software to use E-Cores or P-Cores (and just as well, otherwise all x86 software ever created would need to be revised). It should just work. Plus we don't see these errors on our 12th Gen test system (with E & P cores).

To really get to the bottom of it you would either need some really high end test gear (bus protocol analysers like the Keysight U4970A, etc..) or you do a bunch of trial and error with multiple different RAM sticks and CPUs.