64GB (C)UDIMMs, where are they?

[Chrispy_]

I'm actively trying to get confirmation on this from JEDEC, but I was informed by a vendor that JEDEC 6400 must be CUDIMM. 5600 and below can be UDIMM.

Good news is CUDIMM is backwards compatible, in the sense that if the Motherboard is not configured for CUDIMM it won't see it and acts like normal UDIMM. Downside would be 6400 JEDEC UDIMM is out of specifications if above is true.

Yeah, I found a reddit thread with an AMD employee confirming a few things about bypass mode. I've been unable to find any 64GB regular UDIMMS. In theory if they're making 6400 CUDIMMS there should be 5600, 5200, and 4800 UDIMMS available too but my searches came up empty.

https://www.reddit.com/r/Amd/comments/1gbm20a edit - how do I stop TPU's forum code from insisting that a reddit thread is media to be URL-hijacked, unfurled, and previewed (incorrectly)?

 

[Wirko]

edit - how do I stop TPU's forum code from insisting that a reddit thread is media to be URL-hijacked, unfurled, and previewed (incorrectly)?

I usually make the link part of a paragraph, like igormp did in post #16.

 

[igormp]

FWIW, seems like the combo of 9950x + x670e ProArt does support CUDIMMs in bypass mode:

x.com

x.com

Expected, but seeing an actual example makes me happier with the decision.

 

[Wirko]

I'm actively trying to get confirmation on this from JEDEC, but I was informed by a vendor that JEDEC 6400 must be CUDIMM. 5600 and below can be UDIMM.

Does JEDEC also mean 1.1 volts in all cases? Even in MRDIMMs?

Good news is CUDIMM is backwards compatible, in the sense that if the Motherboard is not configured for CUDIMM it won't see it and acts like normal UDIMM. Downside would be 6400 JEDEC UDIMM is out of specifications if above is true.

Have you done, or are you planning, any compatibility tests? All the early announcements of CUDIMM indeed stated that it should be backwards compatible. And my optimistic expectation was that the clock driver would always have some positive effect because it amplifies and improves the clock signal.

 

[ir_cow]

Expected, but seeing an actual example makes me happier with the decision.

I can show CUDIMM running bypass mode if you like and if you can run 6400 Dual-rank memory, this will be the same but higher density.

Besides actually testing the real kit, that might be the closest you'll get for now.

 

[igormp]

I can show CUDIMM running bypass mode if you like and if you can run 6400 Dual-rank memory, this will be the same but higher density.

Besides actually testing the real kit, that might be the closest you'll get for now.

I'd love to see that, in case it's not a hassle for you!
Otherwise, the already-mentioned reports (specially from @Chrispy_) give me reasonable confidence already, but the 100% certainty mark will only happen once I actually grab the entire setup and build it haha

 

[ir_cow]

Does JEDEC also mean 1.1 volts in all cases? Even in MRDIMMs?

I don't know about MRDIMM but 1.1v JEDEC up to 8800 is in the JEDEC paper.

Have you done, or are you planning, any compatibility tests? All the early announcements of CUDIMM indeed stated that it should be backwards compatible. And my optimistic expectation was that the clock driver would always have some positive effect because it amplifies and improves the clock signal.

Working on article, but it has been delayed for unrelated reasons. I want to see if it works on a bunch of different motherboards as well. Of course the higher speed won't boot because it's beyond what UDIMM can do (maybe the IMC as well). In theory if you had say 6400 CUDIMM, 6400 in bypass mode will work if the Motherboard/ CPU supports 6400 MT/s. That is what I'm testing right now.

The CKD JEDEC papers says unless the clock driver is enabled at startup (by protocols) it will be in bypass mode by default.