Throughout all of this I’ve not seen a clear definition of “EC4” and “EC8”. Do you know where those names come from? Oh, nevermind, found it in a Kingston glossary:
EC4
A JEDEC designation for a DDR5 server-class memory module with a 72-bit data width.
EC8
A JEDEC designation for a DDR5 server-class memory module with an 80-bit data width.
With DDR5 { EC4, x4, x72 } and { EC8, x8, x80 } get used pretty much interchangeably, including in DDR5 JEDEC specs. Just in case it’s not obvious, Kingston’s using server-class to refer to both ECC UDIMMs and RDIMMs, consistent with their x72 definition and common practice in how DIMMs are categorized in product selectors.
It’s a minor distinction but, personally, I prefer EC4 and EC8 when discussing ECC as those are the widths which determine subchannel ECC abilities. x4 and x8 seem easily confused with chip width. x80 isn’t fully interchangeable with EC8 as CB[7:4] are left floating on EC4 RDIMMs.
I wound up going with a 7950X3D and ASRock X670E Taichi. I was very tempted by Threadripper, for the less ambiguous RDIMM ECC situation and what felt like the right number of PCIe slots / lanes, but there’s too much of a gap between $1000~ and > $3000 for me to get into it.
I’ll drop back in with the EDAC output once it gets here.
I would be interested in reading more though… Is that reference publicly available? I wasn’t able to search up the PDF on AMD’s site but maybe I’m using the wrong names for things.
Not sure how deep I can go here without an engineering degree and an industry job.
# dmidecode
...
Handle 0x0010, DMI type 16, 23 bytes
Physical Memory Array
Location: System Board Or Motherboard
Use: System Memory
Error Correction Type: Multi-bit ECC
Maximum Capacity: 128 GB
Error Information Handle: 0x000F
Number Of Devices: 4
...
Handle 0x0015, DMI type 17, 92 bytes
Memory Device
Array Handle: 0x0010
Error Information Handle: 0x0014
Total Width: 72 bits
Data Width: 64 bits
Size: 32 GB
Form Factor: DIMM
Lookit that… I switched CPU and motherboard and 8 of my bits evaporated.
Something else: I was using virtiofs to share a directory with a KVM guest. That had been locking up file access on the VM after a while (somewhere within a 24 hour period of being mounted), leading to hard lockups in any programs that attempted to access the filesystem guest-side. Now, new CPU, that’s not happening anymore.
I’m pretty happy with this, PCIe scarcity notwithstanding. Thanks for all the help, everybody!
(edit) Just for posterity: I retract my remark about the filesystem thing. It’s still locking up on me, just seems to do it less often. Man is that annoying.
I just asked ASRock about their X670e Pro RS model and Ryzen 9 7900 model and they said it can support DDR5 ECC UDIMM module.
Now it seems like on-die ECC has become a marketing thing for DDR5, and it’s hard to understand what RAM is real ECC unregistered or just on-die.
What brand and model provided you with DDR5-6000 ECC UDIMMs ? I’m unable to find any,.
Many thanks
I’m keeping an eye on KSM52E42BD8KM-32HA (KINGSTON 32GB, 5200MT/s, DDR5, ECC, CL42, DIMM, 2Rx8, Hynix A) waiting for the price to lower a bit because I would need two of them. Hopefully two can be used for Dual Channel.