AM5 and DDR5 ECC UDIMMs?

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

unwind-protect

Active Member
Mar 7, 2016
415
156
43
Boston
Since we have the other two threads about the UDIMMs and Alder Lake already:

Any indication whether the DDR5 ECC UDIMMs actually work in current AM5 mainboards?

The 7950X3D coming out tomorrow looks like a good choice, especially from a power standpoint. Just whether the mainboard makers already do or will support ECC with error reporting is a different matter. I don't think ASRock rack has an AM5 board, so IPMI is also hard to get.
 

BigBullion

Member
Jul 28, 2022
45
14
8
I might also try to look at syslog and WHEA and maybe overclock or tighten timings to produce errors like what they did in this article. I am hesitant to overclock or tighten timings because I am afraid that it will damage the RAM or void their warranty.
 
Last edited:

BigBullion

Member
Jul 28, 2022
45
14
8
A forum poster has injected memory errors onto an ASRock B650E ITX board. These errors were shown up in WHEA-Logger. The errors were injected by shorting a data pin to a grind pin on the DIMM. The ASEGA version tested was 1.0.0.5c.

However, even though errors were being reported, it still does not confirm if errors are also being corrected. In order to confirm if it will also correct errors you have to run a memory tester program.
 
Last edited:

BigBullion

Member
Jul 28, 2022
45
14
8
Gigabyte removed the "DDR ECC Configuration" option in their newly updated BIOS. Compare the below screenshots on my Gigabyte B650 Aero G board:

Previous BIOS version:

agesa-1004-ras.JPG

New BIOS version:

agesa-1005-ras.JPG

(This menu can be accessed through: Advanced Mode > Settings > AMD CBS > UMC Common Options > DDR Options > DDR RAS)

------------

I have ran the beta version of MemTest86 that support the AM5 chipset (available here).

Notice that it shows "ECC Enabled: Yes (ECC Correction)".

Previously it only showed "ECC Enabled: N/A (Unknown)".

IMG_6405.JPG

However, according to the below screenshot, something is wrong. It could be that ECC injection is not supported on my board. It could also be that my board corrected the errors silently and did not report them.

IMG_6408.JPG

Yes I did set the "Disable Memory Error Injection" option to "False" is the BIOS.

Conversely, someone with an ASUS X670E motherboard reportedly has ECC injection and/or reporting working. So it might be the case that some motherboard manufacturers support it and some do not.

The below quote asserts that Gigabyte does not report errors even on their AM4.
No, and I think I will never will in regards to x570. Gigabyte's official statement via 1st line support ( the way us mere mortals have to gather information) was something along the lines of: "ecc works by default, no bios settings. Single bit errors are not reported (detected and corrected only) and multi bit errors are not detected and reported.
It is also reported that not just for Gigabyte, all AM4 boards lack official support for ECC reporting. The only reason why some AM4 boards do still report errors might just be because they forgot to disable error reporting. In fact, this is what ASRock acknowledged and they went on to disable error reporting on later BIOS versions. So error reporting is not guaranteed in AM4 consumer grade boards. Even if error reporting is enabled in the current BIOS version, future BIOS versions might disable error reporting.

So many thanks for you detail experience.
We will share this information to RD 

However we got AMD official respond today

* AM4 support ECC function
* AM4 does not support ECC error reporting function

Here is the conclusion:
AM4 platform CPU (Ryzen 1000,2000,3000 series) can all support ECC correction, but not ECC report function

Best regards,
Kevin Hsiueh
Asrock Rack Incorporation
According to AMD, X470 is desktop MB, and our QT won’t test ECC report function on desktop MB.
We follow AMD POR to writes specification.
In order to prevent misunderstanding, we will also remove ”DRAM ECC Error A1/A2/B1/B2” in the IPMI Event Log”.
Thanks for doing so many test and kind remind, and we will pay more attention on similar case in the future.

Best regards,
Kevin Hsiueh
Asrock Rack Incorporation

So it is unknown whether my motherboard corrects memory errors and/or reports them. Testing with Memtest86 shows that it does not appear to support error injection and/or error reporting. Gigabyte consumer-grade AM5 boards also do not have official ECC support (and lack of official error reporting on AM4 as well), and has even gone out of their way to remove the ECC option in their latest BIOS.

It appears unlikely Gigabyte will relist ECC support in their specs later on, due to the fact that they have gone out of their way to remove the ECC option in their latest BIOS.
 
Last edited:

BigBullion

Member
Jul 28, 2022
45
14
8
I think I have confirmed that my Gigabyte B650 Aero G motherboard does not support error injection.

Linux instructions for injecting errors are here: APEI Error INJection — The Linux Kernel documentation

dmesg does not show anything like this:

Code:
ACPI: EINJ 0x000000007370A000 000150 (v01 INTEL           00000001 INTL 00000001)
Also, /sys/firmware/acpi/tables does not contain an EINJ file.

There isn't any WHEA or APEI option in my BIOS.

All those results are with the "Disable Memory Error Injection" set to "False" in BIOS.

Please let me know if you need me to test anything else.
 

Andrew D

New Member
Apr 7, 2023
1
0
1
Sorry for resurrecting this thread, but I've been playing around with the same ASRock ITX board mentioned earlier. I've only checked Linux.

Error reporting is not working, but I suspect that is an issue with the Linux kernel. The amd64_edac module is not checking for family 19h model 61h (Ryzen 7000). On kernel 6.2, this causes the module to fail to load because it is looking for the wrong PCI device IDs. On 6.3-rc5, the module loads because the PCI stuff has been removed for the entire family. However, it defaults to being recognized as generic "F19h", which basically treats it like a Zen 3 EPYC. ras-mc-ctl *seems* to work, but shows 8 channels of memory, and all the DDR5 specific code in the module is being skipped.

Assuming all of the BIOS stuff works (which it might if ECC reporting works in Windows), maybe we can get error reporting with some support in the kernel. I'll also note that the amd64_edac module is looking for Ryzen 5000 (both desktop "Vermeer" F19h M21h and APU "Cezanne" F19h M51h), as well as F19h MA0h, which may be some as-yet-unannounced chip (currently using the same config as Zen 4 EPYC).

I have only looked at EINJ enough to see the BIOS option and notice that the kernel module doesn't load.