ECC UDIMM support on TRX40 motherboards from ASUS, ASRock, Gigabyte, MSI

TL;DR:
What are the launch-day TRX40 motherboards that have real-world support for ECC UDIMMs with single-bit error correction and two-bit error detection?

Dear fellow rippers,

I’m planning on building a personal scientific computer based on a Threadripper 3000-series processor / TRX40 motherboard with ECC UDIMM RAM for tasks of Computational Fluid Dynamics (CFD).

I wonder what kind of experience early adopters are having with true ECC support on the various launch-day motherboards available. I would like to get real-world feedback on the true compatibility of ECC UDIMM modules on TRX40 motherboards.

I’m only interested in “true” ECC functionality here, which means correction of single-bit errors and identification of two-bit errors.
I want to have ECC UDIMMs on a TRX40 motherboard.

Just to make it absolutely clear, I’m not interested in Threadripper TRX40 motherboards only seemingly booting just fine with ECC UDIMMs memory modules installed in DIMM slot but actually running as non-ECC memory. I don’t want to have ECC UDIMMS installed on a TRX40 motherboard and finding out later on that only 8 of the 9 memory chips of ECC UDIMM modules are being used.
I’m also not interested in what the user manuals from manufacturers say, but really interested instead in real-world behavior.

I can’t seem to find relevant ressources online at the time of writing this request (admittedly an early one, only a week after the product launch of the Threadripper 3000-series).

If you respond to this request to kindly share you personal experience, please mention:

  • the make/name of your motherboard and
  • the make/name/product identifier of your memory sticks.

Please also mention which method was used to asses the true support of ECC functionality (or lack thereof), be it with
“wmic MEMORYCHIP get DataWidth,TotalWidth” with TotalWidth 72 on Windows or better with
“wmic memphysical get memoryerrorcorrection” return code on Windows, be it with
"dmidecode --type memory " error correction type on Linux, be it with
memtest86+ or any other meaningful or accurate method.

I hope this thread could end up serving as a resource to people with a similar need.

Cheers!

1 Like

I have been promised an early “real” 2933 kit. I have two sticks I “built” but I want to test quad channel. So far Asus, asrock and gb seeeem to support ecc but still testing.

I usually get that validation by ocing to the ragged edge then monitoring the kernel log.

2 Likes

Thank you, Wendell.
Hope to see a Level1Techs video regarding ECC on Threadripper 3000-series :wink:

1 Like

I’m interested in this too.

Any updates?

NEMIX RAM 128GB 4x32GB DDR4-2666 PC4-21300 2Rx8 ECC Unbuffered Memory is the closest I’ve been able find.
It’s a kit of 4, so would need two kits to get to the full 256 GiB of RAM.

Hello Wendell,

I just put testing the gigabyte trx40 aurus extreme on hold after not being abe to trigger single bit and let alone dual bit errors. Even the most recent build of memtest86 indicates no ecc support.

The methods I used was shorting pins 2 and 5 and also masta killa’s (he has been in contact with you regarding ryzen 9 ecc support) mem tuning settings.

Nothing, nada, zero :frowning: While it is soo easy to do on Ryzen 9 setups where masta killa and I are now able to reliably prove/demonstrate single bit detection, correction and reporting and multi bit detection and reporting on ASUS Prime x570 - P and Asrock Rack x470d4u2-2t (confirmation for x470d4u2 is in progress)

The official statement I got from Gigabyte 1st line tech support, so could be completly BS, is that ECC works by default just no bios settings and that it does not report, only detects and corrects single bit errors. But if we can’t assess that it is we must believe it is not. What engineer in their right mind would leave out the reporting part of ECC? That guy should have become a gardener in my humble opinion.

Anyway, do you or any one following /reading this thread have any results you could share?

I have the Gigabyte TRX40 AORUS MASTER, with 4xMTA18ASF4G72AZ-3G2B1.

It seems to be working in windows I see:

TotalWidth = 128
MemoryErrorCorrection=6

Memetest doesn’t seem that ECC is enabled though with 8.3 Build 1000.

HWInfo say’s ECC is on

Ping the thread for more discussion

I just have SPD timings at the moment but will try to tweek to something a bit faster.

There is 3 BIOS Settings, ECC symbol size, ECC enabled and a 3rd option this caused my computer to reset the BIOS so it assume this isn’t really supported.

The new memtest 9 beta 1, fixes the detection of ECC. Memtest+ doesn’t seem to work, just sort of hangs/ is very unresponsive cf memtest.

MT 9 is out of beta now, so I ran the tests.

It seems like on other gigabyte boards, the ECC injection tests are not supported fully. I do see some that look like they are working and others not. So there is really no concrete way to know if its working, the ECC of single bit errors is maybe invisibly doing it’s job and dual bit errors would be logged.

2021-02-15 16:00:47 - inject_ryzen - UMC_MISCCFG[0] = 00000112
2021-02-15 16:00:47 - inject_ryzen - UMC_ECCERRINJ_0[0] = 00010001
2021-02-15 16:00:47 - inject_ryzen - writing UMC_ECCERRINJ_0[0] = 00010001

Then

2021-02-15 16:00:47 - inject_ryzen - UMC_MISCCFG[2] = 00000117
2021-02-15 16:00:47 - inject_ryzen - UMC error injection configuration writes are disabled (MISCCFG[2] = 00000117)
2021-02-15 16:00:47 - inject_ryzen - writing UMC_MISCCFG[2] = 00000113
2021-02-15 16:00:47 - inject_ryzen - UMC error injection cannot be enabled (MISCCFG[2] = 00000117)

Any updates on this? Has anyone actually got ECC to work properly or followed the steps specified by the original poster and would care to share their results?

I can confirm single bit detection, correction and reporting on gigabyte trx40 aurus extreme rev 1.0

Multi bit I have not gotten around to and probably I never will. I am assuming it is fully supported

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.