ASRock Rack has created the first AM4 socket server boards, X470D4U, X470D4U2-2T

Really appreciate the new BMC firmware, haven’t seen that one before. Is there a changelog?

Hi Folks,

I recently installed two more 16GB UDIMMs on the board.

The X470D4U (probably all X470 boards) will run two UDIMMs at 3200MT without overlocking, and 2666MT with four of them. That is fine.

We see all four ECCDIMMs in the BIOS:

All good? Oh, but wait, Linux has a slightly different opinion:

# dmidecode 3.6
Getting SMBIOS data from sysfs.
SMBIOS 3.3.0 present.

Handle 0x0018, DMI type 16, 23 bytes
Physical Memory Array
        Location: System Board Or Motherboard
        Use: System Memory
        Error Correction Type: None
        Maximum Capacity: 128 GB
        Error Information Handle: 0x0017
        Number Of Devices: 4

I didn’t notice the critical text “Error Correction Type: None” until just now.

Soooo, some bits of information indicate this is a reporting problem, that the DIMMs and the board are indeed error correcting, and that the reporting stops working with 4 DIMMs installed, on… AM4? X470? chipset boards.

I guess I don’t care if it’s really still working. Using Passmark to try and inject ECC errors will also not tell me anything if it’s a reporting issue, so there’s no real way to know I guess.

Any thoughts?

I get the same with only two DIMMs installed:

# dmidecode | rg 'Physical Memory Array' -A10
Physical Memory Array
	Location: System Board Or Motherboard
	Use: System Memory
	Error Correction Type: None
	Maximum Capacity: 128 GB
	Error Information Handle: 0x0017
	Number Of Devices: 4

Sticks are in A1 and B1
image

So I am not sure if 4 sticks is relevant or not. Did you get different output before you added more sticks?

I’ll try to test ECC with BIOS 4.29C and the 3.04.06 IPMI as soon as I got some time, I’ll also try to fetch a Zen 3 CPU, currently using a 3700X.

There has been something broken* for years now how the motherboard reports its memory to an operating system. For example on Windows HWiNFO can’t properly detect the installed memory.

*I’m not exactly sure if it’s genuinely broken or a firmware/software incompatibility. I’ve seen the same on a Gigabyte MC12-LE0, also an AM4 motherboard with IPMI.

1 Like

Unfortunately no updates to report.

1 Like

It’s been a hot minute but IIRC when AM4 launched Wendell was testing ECC on various mainboards and none (or at least few?) reported ECC support, but the error reporting still worked.

It’s probably a quirk of the platform not having “official” ECC support but rather it just… kinda working.

I do wonder if this looks different with a Pro CPU that does have official support.