Installing 24GB DDR5 Modules on AMD Ryzen 7000 Platform Springs Mixed Results—POSTs but Doesn't Boot

[btarunr]

Over the past month, memory manufacturers started releasing DDR5 memory modules of 24 GB and 48 GB densities, which make up 48 GB (2x 24 GB), 96 GB (2x 48 GB or 4x 24 GB) and even 192 GB (4x 48 GB) capacities. There's only one catch—these modules only work with 12th Gen Core "Alder Lake" and 13th Gen Core "Raptor Lake" processors, as their memory controllers support a maximum of 192 GB of memory, and 24/48/96 GB DIMM densities. MEGAsizeGPU decided to find out what happens when one of these kits is installed on an AMD Ryzen 7000 "Zen 4" platform.

A Corsair Vengeance DDR5-5600 48 GB (2x 24 GB) memory kit was installed on a machine consisting of an AMD Ryzen 5 7600X processor, and an ASUS ROG Strix B650E-E Gaming motherboard (BIOS version 1222). It turns out that the machine POSTs, and is able to start the UEFI setup program. Here, the program is able to display the correct 48 GB memory amount, and the memory density of each of the two modules. The trouble is, Windows would not boot, and does not go past the Boot Manager. It halts with an error message that indicates a hardware problem.



AMD Ryzen 7000 series processors technically have a 128 GB maximum memory size limit. Until a couple of months ago, so did 13th Gen and 12th Gen Intel Core processors, which means that this may not be a hard limit, even for AMD, and the company could work with motherboard manufacturers on firmware-level updates to enable support for 24 GB and 48 GB memory modules to at least operate under the 128 GB limit (think 2x 24 GB, 4x 24 GB, or 2x 48 GB). The introduction of 24 GB memory modules improves choices for PC enthusiasts, as it's now possible to have 48 GB of memory using faster single-rank DIMMs (2x 24 GB). It also strikes a middle-ground between 32 GB (2x 16 GB) and 64 GB (2x 32 GB).

View at TechPowerUp Main Site | Source

 

[Chaitanya]

These modules can be very useful for 6 and 12 Core AMD CPUs, just waiting for firmware updates.

 

[konga]

I would consider 48GB as a future-proofed alternative to 32GB since I definitely won't be needing a full 64GB anytime soon. But it sorta needs to work if I were to go that route.

 

[ir_cow]

Memory mapping issues it seems.

 

[LabRat 891]

Memory mapping issues it seems.

It's funny how much things change, how much they stay the same:

Literally went through this same experience trying to find compatible 2GB DDR1 ECC sticks for my K8N-DL. (The 2 GB DDR1 DIMMs were 'planned to be supported' since day1 on the k8n-dl, but examples in the wild didn't really exist until late into the board/platform's existence.)
Even when I finally got a full set+ (that worked very well, only when <8GB installed).... So much 'weirdness', I eventually 'took a long break' from the build. I got 'enthusiastic exhaustion' from trying to get around memory mapping issues.

Fun times. -and now they're back!

 

[stimpy88]

AMD are always months behind on BIOS's and Drivers.

 

[Ferrum Master]

AMD are always months behind on BIOS's and Drivers.

And how AMD is responsible for OEM makers for their lackluster speed of recompiling their BIOS to up to date ALASKA code base? Eh? HOW?

 

[stimpy88]

And how AMD is responsible for OEM makers for their lackluster speed of recompiling their BIOS to up to date ALASKA code base? Eh? HOW?

AMD must have known about the plans for these types of memory modules years ago. So it is their fault that they did not include support for it in their first AGESA release. I remember a video on LTT about 3 months ago where Micron gave Linus a pair of custom built DDR5 memory modules which turned out to be these which are causing AMD issues right now. So AMD must have had these modules at least 6 months ago.

 

[LabRat 891]

And how AMD is responsible for OEM makers for their lackluster speed of recompiling their BIOS to up to date ALASKA code base? Eh? HOW?

Board manufacturers certainly have shown the lack of an industry-standard process and timeline. So, yes, it is on the mobo companies.
Conversely (and I say this as 'generally preferring AMD'), AMD does not provide the same 'level' of devtools Intel typically has*. I mean, it's not like AMD is a fraction of the size of Intel, or anything like that... /s

*This has been at a detriment to the customer before. Intel even provides tools and data packages for board design; sometimes, they're under-engineered designs.

 

[Ferrum Master]

Board manufacturers certainly have shown the lack of an industry-standard process and timeline. So, yes, it is on the mobo companies.
Conversely (and I say this as 'generally preferring AMD'), AMD does not provide the same 'level' of devtools Intel typically has. I mean, it's not like AMD is a fraction of the size of Intel, or anything like that... /s

I can just add, that this problem is not limited only to AMD but intel too. I just fought for months with Gigabyte for a bios to fix MCE errors... So let us be fair.

 

[maxfly]

32, 48, 64, 96 meh. We will survive and so will AMD.

 

[Logoffon]

This is just so vague with bare "failed to boot after multiple tries" error. Are there any boot logs or kernel debugger outputs to pinpoint the issue?

 

[stimpy88]

This is just so vague with bare "failed to boot after multiple tries" error. Are there any boot logs or kernel debugger outputs to pinpoint the issue?

Think BIOS level, not OS...

 

[LabRat 891]

Think BIOS level, not OS...

I don't even know any UEFI commands, but I do know UEFIs are much more featureful and capable than a BIOS. There might be some kind of POST log written to a small protected space in RAM or something.

 

[mrnagant]

What kind of update was pushed for Intel to support these new sized memory modules? Was it more than just a UEFI update? An OS level update? CPU Microcode update?

If UEFI posts, basic compatibility with the motherboard works, especially if memory information is displaying correctly. I am now very curious to know if Linux would work.

 

[marios15]

Board boots
Processor detects RAM properly
BIOS detects RAM size properly
Windows is crashing.

No tests on different OS.

It must be the firmware.





Disclaimer:
yes, it might be the firmware, but 99% of boot issues are caused by Windows behaviour and getting patched in firmware level instead of OS-kernel level is ridiculous

 

[pressing on]

What kind of update was pushed for Intel to support these new sized memory modules? Was it more than just a UEFI update? An OS level update? CPU Microcode update?

If UEFI posts, basic compatibility with the motherboard works, especially if memory information is displaying correctly. I am now very curious to know if Linux would work.

For Asus Z790 boards a BIOS update with the latest ME firmware if not already installed, with predictably only the information that it "Supports high-capacity DDR5 memory kits" so it doesn't seem to involve the OS or CPU microcode. I see that 48GB (2 x 24) and 96GB (2 X 48) Corsair DDR5 memory is already on sale but locally only up to 5600 speed at present.

 

[Mussels]

Nothing new on memory compatibility here - just had to help someone with an x299 system and warnings everywhere about how it wont work with the new single rank 16GB sticks, and that board supports 10th gen CPUs and 512GB total system memory


One platform always has to get support first, since the system POSTs it should be easy to get a BIOS update for compatibility in the long run

AMD must have known about the plans for these types of memory modules years ago.

AMD has been fighting a losing battle against other companies going out of spec for some time, or are you forgetting XMP vs JEDEC?
AMD fought to get EXPO added into the standard for guaranteed compatible RAM, yet a user buys ram with no EXPO or AMD support, not on the motherboard QVL supported list and it doesnt work people are outraged?

Board boots
Processor detects RAM properly
BIOS detects RAM size properly
Windows is crashing.

No tests on different OS.

It must be the firmware.





Disclaimer:
yes, it might be the firmware, but 99% of boot issues are caused by Windows behaviour and getting patched in firmware level instead of OS-kernel level is ridiculous

Can't say you're wrong there, could even be driver related (Thing like an NVME drive using HBM memory having a fit since the memory addresses dont make sense, or a GPU with ReBAR)

 

[robb]

Good grief just spend a little more and get 32GB modules and avoid these issues.

 

[Mussels]

Good grief just spend a little more and get 32GB modules and avoid these issues.

It's fair to make it a news point they aren't compatible yet so users dont get stuck with non-working systems, but some people are taking it, running with it, and trying to become president of "Not understanding this is normal with new RAM technology"

 

[Ferrum Master]

It's fair to make it a news point they aren't compatible yet so users dont get stuck with non-working systems, but some people are taking it, running with it, and trying to become president of "Not understanding this is normal with new RAM technology"

Not sure, but anyone has poked their noses into QVL lists and if the thing is not there, why rising eyebrows if something does not work really. Those lists are there for a reason.

Every board will not work with every RAM as you may think just because of JEDEC. It applies to any CPU, and thanks to OEM makers, as they tend to gimp and screw their board designs. It is not as simple as you all think, but the board is largely involved into RAM init as the clocking is also present on the board not only in the CPU North Bridge.

Reading dumb down stuff and seeing leisure PC expert YT videos does not teach that.

 

[maxfly]

Good grief just spend a little more and get 32GB modules and avoid these issues.

I agree, it's a non issue but it gives us nerds something to ponder...until AMD gets it squared away.

 

[Sora]

This is a ntoskern bsod code (Critical Process Died) correlating to poor ram configuration (Clock/Timings), not a compatibility issue with the dimm configuration on the stick.

A bank compatibility issue would see no post what so ever.

The memory controller capacity printed on intel ark and amd's specs pages are almost never a fact, its just what they could be tested with at the time, westmere for instance taking 48GB despite the 24GB official spec.

 

[Mussels]

I wonder if someone else has tried this yet, a single sample is poor statistics

@ir_cow got your hands on any 24GB modules?

 

[ir_cow]

I wonder if someone else has tried this yet, a single sample is poor statistics

@ir_cow got your hands on any 24GB modules?

I haven't asked Corsair yet for the new 24GB DIMMs, I should though.