Hi!
I felt the same way, when I had to return my RAM kit, which has been bought directly from Nemix RAM.
ECC hasn’t been implemented very well on consumer grade motherboards.
So I’ve bought a Kingston Fury Beast (2x32GB) non-ECC RAM kit, that just works.
I got very lucky with my first Nemix ECC UDIMM kit that was built with Hynix A-dies.
Doubly so to discover my motherboard provides a stable, overclocked AEMP profile that duplicates the speed, timings and performance of my original DDR5-6000 non-ECC memory kit on my daily driver.
Nice to have ECC functionality with no disadvantages compared to standard non-ECC memory kits for once. But I suspect this is a rare case, but appreciate the serendipity.
Hi!
The Nemix RAM kit, that didn’t work properly for me, also had SK Hynix memory chips on the DIMMs.
I am very grateful to the Nemix RAM team, for allowing me to RMA the memory kit.
I’ve bought the kit many months before buying my PC, so the RMA period was over.
Since you can never count on getting any consistently built RAM kits from Nemix, they appear to build with whatever is available on the market at the time.
I will only be using the Kingston kits built with Hynix A-dies specifically mentioned on the product SKU from now on.
Will pay a bit more but at least you know what you are getting when you order it.
Some news guys… I got a non ECC kit from QVL and it didn’t boot either in dual channel…
I took a magnifying glass and saw two bend pins in the socket… Replaced it in the right position with a needle and now it boots perfectly…
I’m infuriating that I sent back two ECC kits which would have probably booted properly… Now I have to buy one more to try again… And lost a lot of euros on shipment fees…
I’ll keep you posted on the results
I have the Asus B650 ProArt as well, and I used four sticks of this
OWC 32GB Replacement for Kingston KSM48E40BD8KM-32HM - DDR5 4800 PC5-38400 CL40 2Rx8 288-pin 1.1V ECC Unbuffered UDIMM Memory RAM Module
the “memory training” took a long time on some of the first boots but once I got past that, there was no issue with these. In the 4x DIMM configuration they run at 3600MHz which is the standard for many / most AM5 CPU’s I think. I did not try to overclock them (and dont plan to), and did not touch the memory settings at all inside BIOS except for enabling ECC. I did a Memtest86 run through and it detected ECC and had no errors. Been running fine for some weeks now. Total memory for the server is now 128GB. Would have been cool to have 192GB but this was good enough for me.
Thanks for sharing this. I bought back the KSM56E46BD8KM-48HM and will try to overclock them, apparently it uses the same hynix-m dies than non ecc sticks.
On another post in this forum, it seems the on huy has bumped the vdd on those to achieve high speed, I’ll keep you posted on the results when I’ll try this
Quick update, the KSM56E46BD8KM-48HM x2 memory kit is booting successfully in dual channel, with 1mn30 of training. So it was an issue of bend pins in the socket…
The board detected an AEMP profile for this kit at 5600 38-38-38-76 at 1.25 V which boots without issue, memtest pro ongoing.
Despite the “Disable Memory Error Injection” BIOS option to false in AMD CBS/DDR Options/DDR RAS (and of course ECC enabled), memtest pro does not succeed at injecting correctable errors.
Lines show [ECC Inject] Injecting ECC error for AMD Ryzen Zen 4 (60h-6fh) but no errors are detected afterwards
Did you disable “Platform First Error Handling” (PFEH)? It should be somewhere under AMD CBS if I remember correctly. With this enabled (usually the default) errors are hidden from the OS (and from memtest86).
Thanks for the tip. I found the parameter in CBS > CPU, but it doesn’t change anything, same behavior than before, no errors correction reported by memtest pro.
I finished my tuning with a satisfactory 6200 Mhz 32 38 38 126, using Buildzoid timings for the rest.
FLCK at 2133 Mhz, temp during stress at 70°C
No errors in Memtest Pro, and also with anta777 extreme profile for memtest5 during the whole night.
It’s nice to see that ECC sticks can perform as well as non ECC ones (which is king of normal since they use the same Hynix dies) !
Does anyone have experience with the Mushkin Redline Pro ECC DIMMs and these (the ProART X870E, X670E, ASrock) boards?
The Mushkin page doesn’t specify much information about the modules… as in, are they Samsung or Hynix, who knows.
I have a 9950x/ProArt X870E with 96GB (48GBx2) Kingston ECC 5600 (KSM56E46BD8KM-48HM) and I had a bit of trouble getting it running. I had to set the memory to 4800 to get it to POST. When I have a bit more time, I’ll try to see if I can tweak things to get closer to 5600.
According to Asus, 4800 is max supported (I can’t post the lnk).
New user here, from googling around this seemed like the thread for AM5 ECC discussion, so I thought I’d chime in with my own two cents.
I bought a 9950X3D to pair with 2x Kingston DDR5-5600 32 GB memory sticks (KSM56E46BD8KM-32HA, uses Hynix A-dies, not in the mobo QVL list) and a X870E Nova motherboard. ECC enabled and PFEH disabled in BIOS as per other comments here. It works and reports errors under Linux with rasdaemon.
After cranking up the frequency and lowering the voltage I was able to generate two corrected errors which were reported correctly by rasdaemon on Arch. It seems to be a delicate balance since the motherboard reverts bad configurations during POST, or the POST proceeds but the system freezes entirely at some point (maybe due to a multi-bit uncorrectable error? No logs are generated that I could look at post-mortem), but after some tweaking I got a configuration that booted to desktop and instantly generated two corrected errors once I started the Blend torture test with mprime.
Running the RAM overclocked at 6000 MHz with the stock CL46 timings seems to be stable, so I’ve started looking for a sweet spot with 1.25 V applied. Full system freezes seem far more common than correctable errors being reported, so I guess when a memory error happens, it corrupts the entire word being transferred? CL40 seemed stable, so I’ve been running stress tests (mprime, stressapptest, Memtest86+) with CL36 timings.
I’m happy so far though; I figured the best scenario was that I’d have to run both 32 GB sticks at stock settings for stability, but seems like there’s far more room for OC than I thought. Or maybe it’s too early to say.
I basically gave up using 4 sticks. Even at stock freq (3600), system freeze is just a matter of time, just take longer than at OC frequencies. The system freeze is especially frustrating as it occurs with no error messages being logged.
Got a 48GBx2 5600 kit from V-Color a few weeks ago and it’s running at the advertised OC freq since then with no issues.
The best way I’ve found to test the ECC reporting on Hynix sticks without too much hassle is to just set the frequency to 7600. At least for me that booted easily but as soon as I tried to run a memory test it’d spit out correctable errors. Earlier I tried super low voltages, but while that worked, finding the right voltage that was just unstable enough was a bit finicky.
My 16GB Kingston sticks are kind of bizarrely good; I eventually settled on running them at DDR5-6400 CL30. That’s kinda nuts! At that point it’s definitely the CPU that’s the limiting factor. I definitely think I lucked out on these sticks, but from what I’ve read about Hynix A-die I think most of it will happily do 6000 CL30 or at least 32 without much trouble. I did notice though that the voltage regulation on these sticks seems locked down; they will not boot with MEM VDD set to 1.425V or higher. Doesn’t matter too much to me, it’s mostly just tCL that scales with VDD and it’s plenty low already.
Below are my settings for reference, but note not all CPU’s will be able to run 6400 in 1:1 mode. For me the SOC voltage seems to be a limiting factor but 1.28V should be safe.
(Edit: apparently the forum software scales down the screenshot making it very hard to read. Here is a full resolution version.)
Regarding the errors I have the same experience, I don’t think I’ve seen a correctable error while tuning this overclock. Of course with ECC you don’t get any memory stress test failures either - it passes or the system freezes, nothing in between.
