Hoping for some availability by the time I’m traveling then
And where art thou, 64 GiB ECC UDIMMs?
48GB ECC UDIMM already has 60GB of memory chips, this is why I guess.
If the 48GB ECC UDIMMs have 60GBs of chips, wouldn’t that make them EC8?
I’m not actually questioning if you are right or not, you are, I’m curious why there’s a seeming consensus on UDIMMs being EC4.
That it would. 54 GB would be EC4.
Basic answer is it’s what the pinout supports.
JEDEC could have kept DDR5 UDIMMs and RDIMMs pin compatible by making UDIMMs x80 or by limiting RDIMMs to x72. Given most RDIMMs seem to be EC4 my guess would be the spec committee’s feeling was EC8 was high enough end it wasn’t worth supporting in UDIMMs. UDIMMs maybe being 5 V and RDIMMs 12 V could also have been a contributing factor to breaking pin compatibility.
but the EC4 RDIMMs have the amount of DRAM packages on them that you would expect for EC4; In the case of a x8 organized dual rank RDIMM, they have 18 DRAM packages, but I’ve got some x8 organized dual rank ECC UDIMMs and they have 20 DRAM packages on them.
I have a hard time believing DRAM module manufactures are putting wasted/dummy dies on their ECC UDIMM modules.
I think your right that it was because JEDEC never made a EC8 UDIMM spec that they are all branded as EC4; but I can’t help but wonder if they are secretly EC8 and just don’t adhere to JEDEC.
…maybe I’m being a little too conspiratorial.
Take a look at the pinouts. RDIMMs have CB[7:0]_A and CB[7:0]_B, UDIMMs CB[3:0]_A and CB[3:0]_B.
That’s definitely what the spec says, but surely those “extra” DRAM packages net out to some of the exposed edge fingers, even if they are just a vss. Intel hasn’t been following JEDEC very well in the recent past (they recently made up there own memory standard that will maybe eventually get codified by JEDEC) so this makes me suspicious.
Looking at the AM5 pinout map on wikimedia, it does appear that there are only 4 bits per 32bit memory subchannel, so I think it’s safe to assume AM5 can only manage EC4… however when I look at LGA1700, it lists a full 8 bits extra per memory subchannel, this kind of implies EC8.
EDIT:
the LGA1851 pinout only specifies 4 bits per 32 bit subchannel implying EC4
I count them on my sticks: 20 chips, each 3GB. Used official SK hynix datashit to determine type of chip. I use Kingston KSM56E46BD8KM-48HM
.
dmidecode -t 17
reports them with 80 bits total width, but Kingston spec clearly says it’s 72 bit. I still don’t know whom I should believe, lol.
Handle 0x0015, DMI type 17, 92 bytes
Memory Device
Array Handle: 0x0013
Error Information Handle: Not Provided
Total Width: 80 bits
Data Width: 64 bits
Size: 48 GB
Form Factor: DIMM
Set: None
Locator: Controller0-ChannelA-DIMM1
Bank Locator: BANK 0
Type: DDR5
Type Detail: Synchronous
Speed: 5600 MT/s
Manufacturer: Kingston
Serial Number: 4204D6B3
Asset Tag: 9876543210
Part Number: 9965794-036.A00G
Rank: 2
Configured Memory Speed: 4800 MT/s
Minimum Voltage: 1.1 V
Maximum Voltage: 1.1 V
Configured Voltage: 1.1 V
Memory Technology: DRAM
Memory Operating Mode Capability: Volatile memory
Firmware Version: Not Specified
Module Manufacturer ID: Bank 2, Hex 0x98
Module Product ID: Unknown
Memory Subsystem Controller Manufacturer ID: Unknown
Memory Subsystem Controller Product ID: Unknown
Non-Volatile Size: None
Volatile Size: 48 GB
Cache Size: None
Logical Size: None
it’s definitely physically 80 bits, like you said with the chip count. but the question is can the platform actually use 80 bits given it isn’t part of the JEDEC spec.
BTW, is that screen from AM5 or LGA1700?
Here’s what it looks like on AM5:
Handle 0x0036, DMI type 17, 92 bytes
Memory Device
Array Handle: 0x0032
Error Information Handle: No Error
Total Width: 72 bits
Data Width: 64 bits
Size: 48 GB
Form Factor: DIMM
Set: None
Locator: DIMMA2
Bank Locator: P0_Node0_Channel0_Dimm1
Type: DDR5
Type Detail: Synchronous Unbuffered (Unregistered)
Speed: 5600 MT/s
Manufacturer: Unknown
Serial Number: 00000000
Asset Tag: DIMMA2_AssetTag (date:00/00)
Part Number: AVW726GUA1K6560P07-H
Rank: 2
Configured Memory Speed: 5200 MT/s
Minimum Voltage: 1.1 V
Maximum Voltage: 1.1 V
Configured Voltage: 1.1 V
Memory Technology: DRAM
Memory Operating Mode Capability: Volatile memory
Firmware Version: Unknown
Module Manufacturer ID: Bank 6, Hex 0xF7
Module Product ID: Unknown
Memory Subsystem Controller Manufacturer ID: Unknown
Memory Subsystem Controller Product ID: Unknown
Non-Volatile Size: None
Volatile Size: 48 GB
Cache Size: None
Logical Size: None
Might be Intel planned the socket so Xeon-E could use RDIMMs but then never went there (the Xeon-E boards I know are UDIMM, anyways).
In my case it’s LGA1700 (W680), ECC option enabled in BIOS.
Seems like this model is finally available in Europe, I found offerings in both DE and UK, hefty price but fast shipping:
Still no sign of those being available in the US tho, Amazon US ships the one from UK.
I haven’t seen any updates from Micron’s side, nor about that other sticks from Kingston (the ones that are more gamery-looking).
All the usual suspects are selling these DIMMs in the US.
Ohhh, that’s great, thank you!
I’m not keen on the usual US stores, apart from Amazon/Microcenter/Newegg, and those other stores often don’t show up on google (at least for me, may be a region thing).
I’ll consider buying those sticks in case nothing else surfaces by next week.
Yeah for the longest time google had provantage blacklisted from product search results and I never understood why, they started to be included recently.
I just want to reiterate for people randomly popping in without reading the whole thread, these are CUDIMMs that don’t work very seamlessly (or at all depending on CPU) with current AMD platforms.
Won’t those work in bypass mode, even if at way lower frequencies?
Theoretically/ideally they should for the 8000/9000 series, but it seems all the different manufactures are handling them differently in BIOS; some will have them boot into the lowest possible jedec speed in bypass mode, some BIOS try to juice the speeds automatically (and possibly cause problems) and some BIOS don’t seem to handle them very well at all.
Perhaps this is just a period of uncertainty in how to implement CUDIMM bypass best and things will get better over time with new BIOS releases.
As an AM5 user - I am quite jealous of these high capacities. 96GB is the highest capacity sweet spot at the moment, which is sadness.