AMD X870E, 9950X3d, and ECC error injection

[MysteryGuy]

Hi;

I now have an Asus X870E Proart motherboard, ECC RAM, and a 9950X3D (Zen 5) CPU.

I ran memtest86 11.5 Pro and I am trying to see if error injection was working with the new CPU and a more recent BIOS since the last time I tried this (hope springs eternal...).

I enabled ECC is BIOS 1715, disabled 'Platform first Error handling', and set 'Disable error injection' to 'false'.

I was pessimistic that enabling ECC error injection would actually work in Memtest (since it didn't work a year ago when I last tried it), but I gave it a shot anyway.

Error injection did not produce any detected errors this time either.

I noticed that

1) I thought it strange that the message show during testing (from memtest) is

[ECC Inject] Injecting error for AMD Ryzen Zen 4 (60h - 6fh)

when the processor is a Zen 5 part as I understand it.

I am under the impression that the memory controller is on the CPU rather than the chipset in this case.

and

2) Similarly, the log file shows

2025-12-14 19:52:45 - [UMC1, CS3] RmSel: 00010000
2025-12-14 19:52:46 - find_mem_controller - AMD Ryzen Zen 4 (60h-6fh) (1022:14E2) at 0-24-2
2025-12-14 19:52:46 - find_mem_controller - AMD Ryzen Zen 4 (60h-6fh) (Ch mode: 2, ECC mode: {detect: yes, correct: yes, scrub: no, chipkill: no, IBECC: no})
2025-12-14 19:52:46 - ECC polling enabled
​

So is memtest using the right technique for a 'Zen 5' type part?

I will attach the log file in case it helps any.

I know that this is a long-shot in hoping that I can get error injection working, but thought I'd give it a try anyway.

Thanks;

[keith]

Thanks for the logs.

Although the logs show Zen 4, it does appear the correct mechanism is being used as per relevant documentation. It is not uncommon for some components within the chipset to be reused across CPU generations.

According to the logs, the relevant registers do not report ECC injection being disabled but the actual registers used for ECC injection appear to be inaccessible. So perhaps there is some undocumented locking mechanism. We'll need to check with AMD.