Return to

16GB ECC UDIMMs in a time of Ryzen 3000

Hi all! I’m based in the UK, and am planning to build a X570/B550-based system soon and am looking at my options for ECC memory. I’ll be starting with 32GB, but may go up to 64GB in time, so I’m looking at 16GB sticks. These generally seem to be Dual Rank, with Single Rank being rare in this capacity.

There are a few options out there, but official 3200Mhz options seem to be just about non-existent - Samsung’s M391A2K43DB1-CWE is still in Sample phase, while Micron’s MTA18ADF2G72AZ-3G2 doesn’t actually seem to be for sale anywhere I can reasonably buy from. The same goes for Kingston’s KSM32ED8/16ME, which uses Micron’s E-die chips. There is Micron MTA9ASF2G72AZ-3G2B1 (Single rank) order-on-request for £98 from a business supplier, but I’m not sure if they’re actually able to get it or how long it might take.

That leaves 2666Mhz options, which a number of people have reported overclocking with varying levels of success.

I can get:

  • Samsung M391A2K43BB1-CTD (B-die) for £99
  • Crucial CT16G4WFD8266-2G6D1 (Haven’t been able to find out what Micron calls this) for £89
  • Kingston KSM26ED8/16ME (Micron E-die, maybe even chips binned for 3200Mhz) for £89

There are one or two Micron Very Low Profile options, but they seem to be 20% more expensive than non-VLP.

Are there any options I’ve missed? How should I go about choosing? Thanks!

Kingston Technology 16GB 3200MHZ DDR4 ECC CL22 DIMM 2RX8 Micron E - $94.54

Are you sure that Samsung still uses B-die on their own modules? Because I heard they stopped and switched to cheaper chips.

Samsung used to have part number decoder guides on their website, including for DDR4:

Unfortunately, I can’t find the DDR4 version archived anywhere, only the DDR3 version, which gives the 10th character as the component revision. Assuming they didn’t change the model number encoding scheme for DDR4, M391A2K43BB1-CTD indicates B-die.

I found and decided to go with Kingston KSM32ED8/16ME, via Probrand, a business supplier. Afterwards, I found that the memory is apparently also available at for £98.

Edit 2020-07-18: Senetic has the Kingston listed for £88, but aren’t currently allowing it to be ordered on their website, only enquiry:

The business supplier I mentioned in my first post is Insight, who seem to accept orders from private individuals on their webstore, and have Micron MTA9ASF2G72AZ-3G2B1 available at for £96 now; the ETA they gave when I phoned in is several weeks.

I found out later that Insight also have Micron MTA18ADF2G72AZ-3G2E1 at for £112, though I suspect it’d also have an ETA of several weeks.

Let us know if it worked. Last year I built a RYZEN system with ECC RAM but there was one type that was wrong and did not work and I had to send it back. I think it has to be ECC unbuffered.

Good research! I am thinking about ordering the Kingston’s myself. Are you going to overclock them? I’m just curious.

I have a Ryzen 1800x currently, but I want to upgrade to Zen3. With the Ryzen 4750G leaks pointing to 2200MHz FLCK clocks, faster memory might be really nice to maximize performance. I don’t know if I want to overclock ecc ram, run ecc ram at stock 3200MHz or buy non-ecc ram.

Off course it is going to work. Unbuffered ECC ram is the ram that works with Ryzen. Just like unbuffered non-ecc. Everything else wont. Registered, Buffered, Fully-buffered and other are server ram.

At the time it was not obvious. ECC RAM never gave me a problem on server hardware but it was not clear what the RYZEN required.

I’ve considered overclocking again (Last time I spent time on overclocking was in the Athlon XP era with the JIUHB chips) and may still do so, but for now I’m running stock.

I’m running the Kingston pair on an X570 Taichi, where HWiNFO picks up that they are ECC, and MemTest86 Pro inidicates that ECC is active. The Taichi BIOS has a setting to enable ECC error injection and MemTest86 Pro reported injecting ECC errors without any problems, but as to whether that actually did anything is unclear. I haven’t tried edac or dmidecode to cross-check the findings, as I’d need to get a LiveCD of some kind running to do that.