Hey all!
I purchased an ASRock B650E PG Riptide because of their advertised “ECC support”. Problem is, they list no qualified ECC UDIMMS in their QVL. ASRock basically recommended I buy and try. But this seems risky from both a validation, and cost standpoint.
Does anyone have this or similar motherboards and know of an ECC UDIMM kit that works?
Thanks!
Asus seems to be the safest bet for ECC on AM5, along with Asrock Rack server/industrial boards. I have an Asus AM5 board but no ECC sticks, but some users on the forum reported working ECC on Asus AM5, use the search function to find their posts.
ECC tests are rare but X670E ASRock’s been done at least twice: Wendell and ApplesOfEpicness. As @quilt mentioned there’s more general Level1Techs threads for this: AM5 ECC and Ryzen 7000.
ECC support’s mostly an AGESA thing, suggesting QVL or ASUS versus ASRock isn’t significant here. So far as I know only Nemix does kits, Kingston, Apacer, and Axiom sell singles. Maybe there’s others.
The physical traces have to be present on the motherboard. ECC dimms have a 72 bit channel, non-ECC have 64. The extra 8 bits have the ECC information.
Some vendors have explicitly no ECC support (I believe MSI?), suggesting the ecc traces are not present so they can never have ECC, whatever you do.
Excellent point regarding the physical traces, however it’s actually 80 bits for ECC DDR5. Gonna steal ServeTheHome’s image:
@quilt is correct. RDIMMs are EC8, ECC UDIMMs are EC4. Specifically, ECC UDIMMs lack RDIMMs’ CB4_A…CB7_A and CB4_B…CB7_B nets.
Not sure if MSI’s ever made an official statement. I know at least as far back as the 1990s Intel’s been giving motherboard manufacturers reference layouts for their sockets. Speaking as someone who used to work with ex-AMD folks in DDR testing, I’d assess it’s almost certain AMD’s doing the same.
It’s possible MSI deleted the CB0_A…CB3_A and CB0_B…CB3_B traces from the AM5 reference but removing CB* drops only four of the 63 signal nets per channel (if I counted the total right). In general I’d expect reference traces to be left in—if it’s not broke, don’t fix it—though it’s possible a ~6% reduction in signal nets from removing check bits crosses a tipping point where a layer can be cost reduced out or something. Haven’t worked with any Gigabyte boards lately but, from what I’ve read, ECC was unofficially possible on AM5 before they pulled the option out of the BIOS.
I suspect really answering this requires knowing people in the relevant product groups well enough they’d be willing to share details and motivations. FWIW, my best guess is MSI and Gigabyte just opted out of showing the BIOS switch because ECC UDIMMs are too niche for them to pick up the test and support cost.
Whoops, thanks for the correction!
Definitely would like to know the answer to the OP’s question myself, a possible Zen 6 drop in upgrade will probably get paired with matching DDR5 if it happens and I’d vastly prefer ECC if that was even an option.
Maybe ASRock’s support can give you a definitive answer.
Yeah I attempted to email AsRock support about this, but I’m not sure they understood the question. After a phone call with their technical support team, they claimed it wasn’t on their QVL because they “couldn’t guarantee access to the memory in American markets” (no I’m not kidding).
Maybe i should buy one stick of DDR5 ECC and give it a whirl; that way worst case scenario I’m only out $80 
The advertising may mean ECC modules work but as non-ECC in the board
I was originally thinking this too, but then I noticed their B650D4U3-2L2Q/BCM (“industrial” AM5 B650 Board) had the exact same spec, and I would be kinda shocked if it didn’t explicitly support ECC. This seems like a bit of a craps hoot, but maybe I should try memory from the QVL on that board (it actually has ECC in it).
I suppose that’s one way to find out, and there are certainly plenty of threads in these forums about how to test and verify that ECC is working. DDR5 with ECC was nonexistent for most of the first year of AM5, so ASRock’s reason checks out. I would be surprised if board vendors bother to spend money updating QVLs for more than a few months post-launch.
Please do follow up if you decide to just nab a stick of ECC and test it yourself!
You’re welcome. The EC4-EC8 SECDED-chipkill distinction’s not well communicated, in my experience, and it’s my understanding the DDR5 spec also defines EC4 RDIMMs. Yet many references I see just say DDR5 does EC8 without even the RDIMM qualification Serve the Home included.
There also seems to be little consideration DDR5 implements write CRC + on-die ECC (~EC2, SEC) + read CRC + ECS (error check and scrub). So non-EC4/8 DDR5 implements much of what’s often assumed to require EC4 and chipkill usually goes unrecognized. PAR breaks out to PAR_{A, B} as well, though that’s minor.
AM5’s well past launch so, personally, I’m fairly comfortable assuming if a mobo allows enabling ECC then AGESA will get EC4 UDIMMs right. It’s not like there’s much concern over CRC retries or whether ECS is running.
Yeah, since you already have the board, just try a stick. ECC sticks should work fine without ECC support too (you just won’t have ECC).
If ECC is a requirement, just get the sticks you wanted and swap out the board if it doesn’t work…
