Wednesday, November 18th 2026

JEDEC Previews Next-Generation DDR6 DRAM

JEDEC Solid State Technology Association, the global leader in the development of standards for the microelectronics industry, today unveiled a preview of its DDR6 DRAM standard, the next-generation main memory standard designed to support the accelerating demands of AI, cloud, data center, server and client computing. Currently in development in JEDEC, DDR6 will introduce a new architecture aimed at delivering significant gains in bandwidth, capacity, power efficiency and reliability, helping to enable the next era of high-performance computing.

The following features reflect current JEDEC DDR6 development plans rather than a finalized standard, and are subject to change through JEDEC's development and approval process:
Increased Performance
  • Higher bandwidth: Improved data rates and a wider bus will combine to increase bandwidth more than 50% at introduction, with further bandwidth increases as data rates increase. This bandwidth target is distinct from a claim of equivalent improvement in application performance.
  • Higher data rates: DDR6 RDIMMs move beyond DDR5's 9.2 GT/s to a data rate that will start at 12.8 GT/s with a roadmap extending to 14.4 GT/s and beyond.
  • Wider bus: The data bus expands from 64 to 96 DQ, a 50% increase in data bus width.
  • Pseudo-channel architecture: DDR6's planned architecture organizes memory access into pseudo-channels to support higher bandwidth per DRAM package. It is intended to provide greater configuration flexibility for system reliability and more efficient data movement.
  • Higher bank availability: 2X bank count and improved refresh boost data bus efficiency, enabling system performance at higher speed grades and higher capacities.
Higher Capacities
  • Higher DRAM density: Support for up to 128 Gb DRAM die density.
  • Efficient capacity scaling: Optimizations for die stacking in DDR6 packages delivers more density per package footprint.
  • Compact module footprint: A significantly narrower 74 mm DDR6 module length, compared with 133 mm for DDR5, is intended to help accommodate higher system memory density.
Reliability
  • Flexible reliability: Support for multiple system error-correcting code (ECC) implementations will allow system designers to balance cost, power, and reliability at the system level.
  • Built for the data center: Interface optimized for high-capacity modules with configurations oriented to general purpose compute as well as AI inferencing and agentic AI, which demand more capacity and performance close to the CPU.
Energy Efficiency
  • Improved performance-per-watt: DDR6 targets significantly improved performance-per-watt versus DDR5 through a change to VSS-terminated IO and data-center-focused low power modes.
Form Factors
  • Breadth of form factors: DDR6 is being defined across multiple form factors—potentially including RDIMM, UDIMM, and CAMM to serve a broad range of applications while balancing reliability, performance, power, and cost.
Christopher Cox, Chair of the JC-42 Committee for Solid State Memories"As AI and data-intensive workloads continue to drive system requirements higher, DDR6 is being designed to deliver meaningful advances in memory bandwidth, capacity, and reliability. The collaboration of JEDEC member companies is helping create a next-generation memory standard that balances performance, power efficiency, and scalability across a broad range of computing applications."
Mian Quddus, JEDEC Board of Directors Chairman"JEDEC members gain early access to pre-publication proposals and insights into active projects such as DDR6. We welcome interested companies worldwide to join JEDEC today to help shape the future of standards for this dynamic industry."
Add your own comment

25 Comments on JEDEC Previews Next-Generation DDR6 DRAM

#1
Onasi
Huh, so they decided not to go CAMM-only after all, interesting. I guess it’s a data center driven decision?
Posted on Reply
#2
enb141
I think the problem with CAMM-only is that servers and desktops doesn't seem to get benefit that laptops can get with CAMM.

Now, when the final DDR6 will be released?
Posted on Reply
#3
r00tcz
enb141I think the problem with CAMM-only is that servers and desktops doesn't seem to get benefit that laptops can get with CAMM.
Huh, so they decided not to go CAMM-only after all, interesting. I guess it’s a data center driven decision?
Looks like it's SODIMMs sized modules only:
Compact module footprint: A significantly narrower 74 mm DDR6 module length, compared with 133 mm for DDR5, is intended to help accommodate higher system memory density.
Not as good as CAMM but still improves signal integrity a bit, not having to route signals across 13cm long connector.
Posted on Reply
#4
Onasi
enb141I think the problem with CAMM-only is that servers and desktops doesn't seem to get benefit that laptops can get with CAMM.
Eh, theoretically, signal integrity and no obtrusion for cooling are still good benefits.
Posted on Reply
#5
Mysteoa
Would this finally make integrated GPU not starve for bandwidth? It looks like it can reach around 300Gb/s for dualchanel, closing on low clocked GDDR6
Posted on Reply
#6
Hecate91
I don't care if DDR6 is going to be CAMM or DIMM, it's going to be hilariously expensive anyway unless the AI bubble pops by the then.
Posted on Reply
#7
evernessince
OnasiHuh, so they decided not to go CAMM-only after all, interesting. I guess it’s a data center driven decision?
CUDIMM solved the signal integrity issue memory was starting to have on desktops, for now at least. The only other benefit of CAMM was it's lower height, which doesn't really benefit desktop much. We might seen it at a later time though if signal integrity becomes an issue again.
Posted on Reply
#8
xxxZAURxxx
3 days later, JEDEC
All production for 2027 has been reserved.
Posted on Reply
#9
Onasi
@evernessince
I don’t remember off the top of my head, but does CAMM also need a CKD to reach higher frequencies like conventional DIMMs? If not, that’s theoretically an advantage on costs. If yes, yeah, then you have a good point.
Posted on Reply
#10
xxxZAURxxx
Unless you actually sell these technologies to users, it doesn’t matter if you go ahead and produce DDR16 and pat yourself on the back—nothing changes.

I’m currently using DDR12 myself. But who cares?
Posted on Reply
#11
windwhirl
xxxZAURxxxUnless you actually sell these technologies to users, it doesn’t matter if you go ahead and produce DDR16 and pat yourself on the back—nothing changes.

I’m currently using DDR12 myself. But who cares?
Guess what, the users are now the enterprise sector. Mission accomplished.
Posted on Reply
#12
Athena
Hecate91I don't care if DDR6 is going to be CAMM or DIMM, it's going to be hilariously expensive anyway unless the AI bubble pops by the then.
I would say it still would be expensive whether or no the AI bubble pops
The retooling alone is billions of $
Posted on Reply
#13
xxxZAURxxx
windwhirlGuess what, the users are now the enterprise sector. Mission accomplished.
In that case, the DDR6 announcement is a complete lie as far as I’m concerned.

Let them play their own music and dance to their own tune.
Posted on Reply
#14
NeonSky77
xxxZAURxxxUnless you actually sell these technologies to users, it doesn’t matter if you go ahead and produce DDR16 and pat yourself on the back—nothing changes.

I’m currently using DDR12 myself. But who cares?
you from the future?
Posted on Reply
#15
wNotyarD
Onasi@evernessince
I don’t remember off the top of my head, but does CAMM also need a CKD to reach higher frequencies like conventional DIMMs? If not, that’s theoretically an advantage on costs. If yes, yeah, then you have a good point.
The JEDEC standard for CKD encompasses CAMM applications. How much does CAMM needs a CKD in comparison to DIMM is anybody's guess.
Posted on Reply
#16
ir_cow
Onasi@evernessince
I don’t remember off the top of my head, but does CAMM also need a CKD to reach higher frequencies like conventional DIMMs? If not, that’s theoretically an advantage on costs. If yes, yeah, then you have a good point.
It's optional for the most part to reach 9600. Beyond that I think it's CKD only for signal integrity
Posted on Reply
#17
jesdals
AI AI lets get to DDR6 soon so us mere mortals can get affordable DDR5 again...
Posted on Reply
#18
Chomiq
jesdalsAI AI lets get to DDR6 soon so us mere mortals can get affordable DDR5 again...
It's still space on a wafer.
Posted on Reply
#19
Hecate91
AthenaI would say it still would be expensive whether or no the AI bubble pops
The retooling alone is billions of $
I'm expecting it'll be more expensive than DDR5 regardless of an AI bubble pop.
DDR5 was very expensive at launch, in the current market it's more expensive than it ever was, and I expect DDR6 will be more with the excuse that it's faster.
xxxZAURxxxUnless you actually sell these technologies to users, it doesn’t matter if you go ahead and produce DDR16 and pat yourself on the back—nothing changes.

I’m currently using DDR12 myself. But who cares?
DDR12? Are you from the future?
Posted on Reply
#20
thestryker6
A lot of interesting stuff there even if we're unlikely to see it this decade the way things are going. Seems like this ought to mean the death of SODIMMs since the module length they're talking isn't much more than current SODIMMs. I'm curious if there will be any latency or thermal implications from the area reduction. I believe this is the first time that the bus width increase has been confirmed though it was always likely since LPDDR6 had increased. A starting bandwidth of, at least on paper, ~307GB/s is an increase of around double the best we've got today. The implications for integrated graphics and high core count desktop parts are pretty big.

I think the specification was supposed to have been released this year and clearly isn't given this announcement. The only thing I can think of is that given the mempocalypse they're kicking the can down the road and it will be next year. With DDR5 we saw it on client first, but I'd be surprised if that happened with DDR6 even if validation for enterprise takes longer. Really unfortunate for something that stands to be the biggest change for the memory market since dual channel.
Posted on Reply
#21
TumbleGeorge
I can't hide that I'm glad that the people who bet and hoped that camm2 would be imposed almost by force on the market are now losing.
Posted on Reply
#22
Athena
TumbleGeorgeI can't hide that I'm glad that the people who bet and hoped that camm2 would be imposed almost by force on the market are now losing.
Why? I mean, CAMM2 is better for laptops, for desktop, it does have some benefits, more clearance, better signal integrity , which, was mentioned above
Posted on Reply
#24
Hecate91
AthenaWhy? I mean, CAMM2 is better for laptops, for desktop, it does have some benefits, more clearance, better signal integrity , which, was mentioned above
CAMM2 only offers a single memory socket limiting upgrade capability, on a significantly more complicated fragile LGA socket.
Posted on Reply
#25
evernessince
Onasi@evernessince
I don’t remember off the top of my head, but does CAMM also need a CKD to reach higher frequencies like conventional DIMMs? If not, that’s theoretically an advantage on costs. If yes, yeah, then you have a good point.
It doesn't require it but JEDEC spec recommends it above 6400. JEDEC is very conservative. Most kits can hit way above that without a CKD.
Posted on Reply
Add your own comment
Nov 18th, 2026 16:53 PST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts