Friday, April 24th 2026

AMD Launches EXPO 1.2 for CUDIMM and Low-Latency DDR5 Memory

by
AleksandarK
 Discuss (20 Comments)
AMD has officially launched its Extended Profiles for Overclocking (EXPO) version 1.2, which expands support for more memory profiles, CUDIMM, and additional features. With this release, AMD aims to introduce more advanced memory tuning profiles to the "Zen 5" platform and its DDR5 memory types, while full support for EXPO 1.2 features is anticipated with the "Zen 6" processor generation. This means that full CUDIMM support for "Zen 5" tuning is still limited, and only "Zen 6" will offer complete CUDIMM support. EXPO 1.2 works with AMD's AGESA, but since AGESA 1.3.0.1 lacks full CUDIMM DDR5 support, this feature is postponed until the next-generation "Zen 6." Current-generation "Zen 5" CPUs on the AM5 socket may be able to use EXPO 1.2, but they won't fully support the new DDR5 memory with an onboard clock driver, instead relying on traditional memory to perform the task. Notably, popular AMD developer 1usmus mentioned that AMD EXPO 1.2 brings the following changes.
  • Support for module geometry.
  • Added support for MRDIMM (CUDIMM and CSODIMM are already supported). AGESA 1.3.0.1 still lacks full CUDIMM support; AMD is preparing this for Zen 6.
  • New! tREFI, tRRDS, tWR, ULL Enable—Unified Latency Lock and VDDP(V).
Finally, AMD is adding Chinese memory makers to EXPO 1.2 as DRAM shortages are prompting customers to seek alternatives to traditional consumer DRAM kits. These include brands such as RAMXEED Limited Conexant, Rui Xuan (formerly known as Rei Zuan), and Fujitsu Synaptics. AMD EXPO 1.2 is now being rolled out to some existing X870 motherboards for "Zen 5," although CUDIMM support is only partial, and details about it are still limited. These boards include the ASUS ROG CROSSHAIR/STRIX/PROART/TUF X870 Series with beta BIOS version 2301.
Sources: 1usmus on X, via VideoCardz

Related News

Add your own comment

20 Comments on AMD Launches EXPO 1.2 for CUDIMM and Low-Latency DDR5 Memory

#1
Onasi
C'mon AMD, you are saving money by NOT proof-reading public facing official slides now? "Doucments", seriously? Any auto-correct would catch that. It's a petty nitpick, sure, but this is just mildly embarassing.
Posted on Reply
#2
Athena
CUDIMMs are even more expensive, but, I'd think these would be great for o/c'ing diehards
Posted on Reply
#3
dgianstefani
TPU Proofreader
OnasiC'mon AMD, you are saving money by NOT proof-reading public facing official slides now? "Doucments", seriously? Any auto-correct would catch that. It's a petty nitpick, sure, but this is just mildly embarassing.
It's about as production ready as AGESA code.
Posted on Reply
#4
Chaitanya
Interesting addition, hopefully Zen6 will atleast improve memory training timings seen on current AM5 CPUs.
Posted on Reply
#5
_roman_
Less overclocking and insane timings means faster am5 boottimes most likely. are up to ~20 seconds even worth discuss about?
Posted on Reply
#6
Jtuck9
OnasiC'mon AMD, you are saving money by NOT proof-reading public facing official slides now? "Doucments", seriously? Any auto-correct would catch that. It's a petty nitpick, sure, but this is just mildly embarassing.
If only the majority of us had the luxury of "failing upwards"...
Posted on Reply
#7
bonehead123
_roman_Less overclocking and insane timings means faster am5 boottimes most likely. are up to ~20 seconds even worth discuss about?
Well, my rig already boots faster that than, and it's several years old, running that 645GB of that old-school stuff called DDR4 SO-DIMMS, hahaha :)
Posted on Reply
#8
igormp
AthenaCUDIMMs are even more expensive, but, I'd think these would be great for o/c'ing diehards
I have CUDIMMs (4x64GB 6400MHz) with my current setup in bypass mode, and any kind of extra support for those would be awesome to achieve higher frequencies, although the current 5600MHz they're running at is more than I had expected out of them.
Posted on Reply
#9
jesdals
More support for more suppliers - whats not to like - a demand for a new chipset to be able to run it?
Posted on Reply
#10
_roman_
the memory controller is afaik in the cpu. when amd did not botch it, it is only a firmware issue.

i really wonder if cudimm is an intel invention? does it follow the standard? or does cudimm violate what jedec or whatever made ddr5 ? when it needs extra datalines you will need a new motherboard with most likey a rebranded or the same chipset
Posted on Reply
#11
AusWolf
ChaitanyaInteresting addition, hopefully Zen6 will atleast improve memory training timings seen on current AM5 CPUs.
I hope so too. I'm also hoping for more sane levels of SOC voltage with 6000-6400 MHz RAM. Throw stones at me, but I do think that 1.2 V is too much when you're defaulting to 1 V with 4800 MHz JEDEC. You definitely feel the difference in CPU temperature and power consumption.
Posted on Reply
#12
_roman_
I think you mix something up. there is the direct current voltage on the DDR5 DIMs themselves which are downstepped from the 5V DC supply on the dimm and there is the voltage for the memory on the cpu side. AFAIK MSI changes 3 values, which is very bad at the same time to 1.35V. I set two values identical and one cpu value much lower.

Those dimm values are more an indicator for a bad bin of the DRAM modules. i bought those dram modules on purpose as 1.25V DC was kinda lower as the other similar modules at that time KF556C36-32. I would not touch 1.40 or 1.45V DC for 6000 mt/s or similar.
Posted on Reply
#13
AusWolf
_roman_I think you mix something up. there is the direct current voltage on the DDR5 DIMs themselves which are downstepped from the 5V DC supply on the dimm and there is the voltage for the memory on the cpu side. AFAIK MSI changes 3 values, which is very bad at the same time to 1.35V. I set two values identical and one cpu value much lower.

Those dimm values are more an indicator for a bad bin of the DRAM modules. i bought those dram modules on purpose as 1.25V DC was kinda lower as the other similar modules at that time KF556C36-32. I would not touch 1.40 or 1.45V DC for 6000 mt/s or similar.
I'm not talking about the DRAM voltage. I'm talking about the SOC voltage needed to run your memory controller, and as a result, your RAM at specific speeds.

6000 MT/s RAM needs 3000 MHz on the memory controller, which needs anywhere between 1.2-1.3 VSOC on current AM5 CPUs to be stable.
4800 MT/s RAM works with a 2400 MHz memory controller, which defaults to 1 VSOC on my board, which lowers CPU power consumption by 10 W, which can be important when you're power or thermal limited.
Posted on Reply
#14
Pooch
Why is 1080 still even a thing anymore? 1080 was good when it was the highest option. Now, compared to 1440 and 4k, it is crap. My point it that 1080 should no longer be mentioned as a measure for performance. 1080 is what made the switch look like garbage, if they would have employed some sort of antialiasing maybe ok, but the switch barely had enough power for 1080 without it. Lets leave 1080 behind.
Posted on Reply
#15
tommo1982
PoochWhy is 1080 still even a thing anymore? 1080 was good when it was the highest option. Now, compared to 1440 and 4k, it is crap. My point it that 1080 should no longer be mentioned as a measure for performance. 1080 is what made the switch look like garbage, if they would have employed some sort of antialiasing maybe ok, but the switch barely had enough power for 1080 without it. Lets leave 1080 behind.
1080p is relevant. I play at this resolution, simply due to cost of high end GPU's and their power consumption. That and I have a good 1080p TN monitor which I bought for pennies and replacing it with something similar would be costly.
Posted on Reply
#16
AusWolf
PoochWhy is 1080 still even a thing anymore? 1080 was good when it was the highest option. Now, compared to 1440 and 4k, it is crap. My point it that 1080 should no longer be mentioned as a measure for performance. 1080 is what made the switch look like garbage, if they would have employed some sort of antialiasing maybe ok, but the switch barely had enough power for 1080 without it. Lets leave 1080 behind.
Resolutions above 1080 are in diminishing returns territory in terms of visual quality and the GPU power required to run it (says someone who games on 1440 ultrawide).
Posted on Reply
#17
_roman_
target amd gpu windows 11 pro driver is set to veritical 45hz freesync / whqd :) is my favourite

i'm not sure. i may have misunderstood. partly cudimm support will be available soon for any am5 cpu. full implementation needs a newer silicion as i suspect intel did something on their own again. high mt/s but worse latency of a redriver cirucit. some want cudimm. i do not want cudimm because of the redriver circuitry.
Posted on Reply
#18
ir_cow
Adding tREFi is dangerous PR stunt. I can see the RMA requests rolling in when memory errors out above 40c with 65k trefi. Average consumer has no idea the reason, just the system crashes lol.
_roman_the memory controller is afaik in the cpu. when amd did not botch it, it is only a firmware issue.

i really wonder if cudimm is an intel invention? does it follow the standard? or does cudimm violate what jedec or whatever made ddr5 ? when it needs extra datalines you will need a new motherboard with most likey a rebranded or the same chipset
CUDIMM isn't a Intel invention like XMP. Standard comes from the JEDEC group. AMD can't run clk because it was never tested for zen4/5 during development. The cost to validate pre existing CPUs and chipset is too great. AMD would only do it if they were super behind in performance. Considering x3D is still top with just 6000 memory, I don't think AMD is in a rush to add backwards support for CUDIMM.
AthenaCUDIMMs are even more expensive, but, I'd think these would be great for o/c'ing diehards
Yeah but CAMM2 is becoming the new standard with CLK for higher OC and JEDEC officially only recognizes 6400 and above as CUDIMM. sooo only a matter of time before AMD has to adopt it as well.
Posted on Reply
#19
R-T-B
_roman_the memory controller is afaik in the cpu.
Not on AM5 no.
Posted on Reply
#20
rv8000
ir_cowAdding tREFi is dangerous PR stunt. I can see the RMA requests rolling in when memory errors out above 40c with 65k trefi. Average consumer has no idea the reason, just the system crashes lol.


CUDIMM isn't a Intel invention like XMP. Standard comes from the JEDEC group. AMD can't run clk because it was never tested for zen4/5 during development. The cost to validate pre existing CPUs and chipset is too great. AMD would only do it if they were super behind in performance. Considering x3D is still top with just 6000 memory, I don't think AMD is in a rush to add backwards support for CUDIMM.


Yeah but CAMM2 is becoming the new standard with CLK for higher OC and JEDEC officially only recognizes 6400 and above as CUDIMM. sooo only a matter of time before AMD has to adopt it as well.
Ive run 16-24 hour tests with max trefi while in the 48-52c range from hwinfo (not even the actual temp) and had zero issues.

The various camm standards also have no purpose for consumer desktop pcs, I have my doubts we’ll even see an implementation thats more than just a one off during AM6.
Posted on Reply
Add your own comment
Nov 12th, 2026 04:42 CST change timezone

Latest GPU Drivers

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts