Validating ECC RAM on AM4 Ryzen Consumer Motherboard

I’m in the process of switching motherboard from ASRock B550M-HDV to an Asus ProArt B550-CREATOR and thought it would be interesting to do this test on both the old and new motherboards. So far I’ve been having a real struggle with the old board:

  • CPU: Ryzen PRO 4650G (widely reported working ECC support)
  • Motherboard: ASRock B550M-HDV (single slot per channel!)
    • UEFI: 3.40 (AGESA 1.2.0.B)
  • RAM: 2x Mushkin 16 GiB (MPL4E320NF16G18)

Default settings for these DIMMs are 3200 MT/s, 22-22-22-22-52, 1.20 V.

This is what I’ve tried:

(“OK” here means boots and runs memtest86 v11.2 for at least 5 minutes without any errors detected. “FAIL” = fails to boot; just a black screen, no beep codes.)

19-21-21-21-46: OK
18-20-20-20-45: OK
17-20-20-20-42: OK
16-20-20-20-40: FAIL
16-20-20-20-42: FAIL
17-20-20-20-39: OK
17-19-19-19-39: OK
17-19-19-19-36: OK
17-18-18-18-36: Crash to black screen in memtest within seconds
17-18-19-19-36: Crash to black screen in memtest within seconds
17-19-18-18-36: OK (ran memtest86 for 45 minutes)
17-19-18-18-33: OK
17-19-18-18-33 @ 1.18V: OK
17-19-18-18-33 @ 1.16V: OK
17-19-18-18-33 @ 1.14V: OK
17-19-18-18-33 @ 1.12V: Fails to boot, 3 long beeps
17-19-18-18-33 @ 1.13V: Crashes within seconds in memtest86 and UEFI

Ok, so change of tactic: What I want is marginal signalling on data transfers and not marginal function of command signals, so let’s use the default primary timings and lower voltage and increase frequency instead:

Timings set to Auto (so clock counts adjusted to keep actual timings constant with increasing frequency):

3200 MT/s @ 1.13 V: OK (ran memtest86 for 20 minutes)
3200 MT/s @ 1.12 V: OK
3200 MT/s @ 1.10 V: OK
(3200 MT/s @ 1.08 V: Nope, UEFI won’t let me lower voltage this far)
3400 MT/s @ 1.10 V: OK (ran for 15 minutes)
3600 MT/s @ 1.10 V: OK (ran for 25 minutes)
3800 MT/s @ 1.10 V: OK (ran for 10 minutes)
4000 MT/s @ 1.10 V: FAIL
3933 MT/s @ 1.10 V: Edit: Ran for almost three passes without any errors, then crashed. :frowning: So I think this config is very much on the verge of instability and should have produced ECC errors if the reporting was working.

All this without a single reported ECC error. ECC is enabled and memtest86 is reporting ECC support and ECC polling enabled. However there is no “Platform First Error Handling (PFEH)” setting in the UEFI. I’m obviously at this point wondering if reporting really is disabled on this motherboard.

Any thoughts?

Edit:

Here’s one of the previous (unfortunately now closed) threads on the same subject: Ryzen 5700X ECC reporting.

Notifying some of the most active members from that thread: @aBav.Normie-Pleb @brain-short @fornex