Get ECC uncorrectable errors on Alder Lake platform.

[Tank]

I've noticed that when I only insert one ECC SODIMM into the B1 slot (the one farthest from the CPU), I get ECC uncorrectable errors. But when I insert it into the A1 slot (the one closest to the CPU), I don't get any errors. What could be the reason for this?
I also upload memtest log attachment for reference.Thanks.

Platform: Alder Lake(1DPC)
CPU:Intel Core i5-12500E
Chipset: R680E
Memtest version: V11.3 Pro build 1000

[David (PassMark)]

I can see in the log that you enabled ECC injection, for at least part of the time. So maybe the errors in the screen shot are expected?

Otherwise differences in behavior in different slots can be due to longer (or shorter) track lengths on the PCB changing the electrical and timing characteristics for the slot, for example higher voltage drop. Or dirt on the connector. Or EMI. Or temperature / airflow differences.

[Tank]

No, I don't have enabled ECC injection, just only execute standard test. You can see screenshot a lot of ECC errors after 15 seconds.

I wonder to know what could be the problem with the ECC Uncorrectable error, I'm using an ECC DIMM which should be corrected the error. Thanks.

[keith]

This appears to be a false positive when accessing the ECC registers on unpopulated slots. Please check your messages for a link to an updated build.

[codingfou]

Hello !
I have the exact same issue but on an Intel N100 (Alder Lake N platform)


On this computer, I use In Band ECC which is an ECC method using actual memory instead of ECC sticks.

If I disable In Band ECC in the BIOS : no error
With In Band ECC enabled in the BIOS and memtest86+ : no error

Given the error is always at the same address, and at the top upper bound, I wonder if it's an issue in the ECC polling of memtest86.

Apparently, the developers of the Linux EDAC driver (driver for ECC error monitoring) encountered the same issue on this platform !
Polling the ECC register would return a value of ~0 (0xffffffff) and flood error messages

This really seems that's what happen in memtest86 (especially given the memory address at the upper bound !)

They solved this by clearing the register.

You can see the git diff on the driver here : https://github.com/torvalds/linux/co...b6f15aca57da92

This would be very nice to fix that !

[David (PassMark)]

Yes, that is what the MemTest86 log showed. All F's. Which resulted in false positives on the unpopulated memory slots.
ECC RAM has been around for 50 years. It is pure insanity that Intel and the other CPU vendors still can't agree on a standard for handling ECC RAM and new code is required to be written for each CPU family.

codingfou, I've sent you a private message in the forums here, with a link to a new built to try (we don't have a N100 CPU to test with).

[codingfou]

Hello
I have tried the new build today and it fix the issue
Thanks !