Friday, February 20th 2026
AMD Next-Gen Ryzen "Olympic Ridge" Zen 6 Delayed to 2027
AMD's next-generation Ryzen "Olympic Ridge" desktop processors are rumored to have been delayed to 2027. Company roadmap slides point to a debut of the "Zen 6" microarchitecture in 2026. This could be a signal that AMD is prioritizing "Zen 6" chiplets for EPYC server processors as the enterprise market has a better-secured supply of memory, and enterprise customers are more likely to stick to their procurement roadmaps for CPUs, in stark contrast to the client segment, which has been grappling with acute shortages of DDR5 memory.
With PC memory selling at prices 5-8 times above normal, 2026 may not present an ideal environment for AMD to debut a new Ryzen desktop processor generation, especially given that AMD processors tend to be heavily favored by the DIY retail channel compared to Intel, which retains a hold over the pre-built OEM market. Besides increased IPC and an updated ISA, "Zen 6" is expected to see AMD increase CPU core counts, as well as update the client I/O die. This new cIOD, likely built on 4 nm, is expected to come with updated DDR5 memory controllers that support higher speeds and have a greater degree of parallelism for dual-channel DDR5, likely with each controller addressing two 40-bit sub-channels from different channels, an approach similar to Intel's with "Arrow Lake." If your best feature is, say, DDR5-9000 support with CUDIMM optimization, you rather hold off when a 32 GB >9000 MT/s CUDIMM kit costs upward of $800.
Sources:
Benchlife.info, VideoCardz
With PC memory selling at prices 5-8 times above normal, 2026 may not present an ideal environment for AMD to debut a new Ryzen desktop processor generation, especially given that AMD processors tend to be heavily favored by the DIY retail channel compared to Intel, which retains a hold over the pre-built OEM market. Besides increased IPC and an updated ISA, "Zen 6" is expected to see AMD increase CPU core counts, as well as update the client I/O die. This new cIOD, likely built on 4 nm, is expected to come with updated DDR5 memory controllers that support higher speeds and have a greater degree of parallelism for dual-channel DDR5, likely with each controller addressing two 40-bit sub-channels from different channels, an approach similar to Intel's with "Arrow Lake." If your best feature is, say, DDR5-9000 support with CUDIMM optimization, you rather hold off when a 32 GB >9000 MT/s CUDIMM kit costs upward of $800.
87 Comments on AMD Next-Gen Ryzen "Olympic Ridge" Zen 6 Delayed to 2027
Meanwhile gamers cope and seethe that Apple is shipping these parts every year. All they can buy is garbage because AMD/Intel are lazy/incompetent respectively.
Remember: Zen 3 was the last time AMD shipped anything on time. Ever since then they've delayed their parts for various reasons. Zen 4 it was to add CXL support (you don't care about that, no consumer does - yet it delayed shipping by 4 months). Zen 5 it was because they scrapped their N3B plans early and switched to N4P/N3E to save themselves money but it delayed consumer parts about 3 months.
Zen 6 consumer parts being delayed about 3-4 months (to 2027Q1) seems in line with past AMD laziness.
And all this cumulative delay means that the part that should be nearly as fast as M5 will be shipping more than a year after it. Zen 3 was the last time they shipped on time and was also the last time they were competitive in SPECint 2017 1T.
nothing to throw a wrench in the 2027 launch of ddr6 for servers
lpddr6 for mobile is ready now.Very few people upgrade their systems.
AMD doesn't make enough money on them to waste time on.
And restarting AM4 production is out of the question. That would require new agreements with TSMC and Global Foundries, and those legacy nodes (7nm and 12nm, respectively) are probably already sold out for years for automotive, industrial, etc.
Although it wouldn't surprise me, every company is fully focused on AI with no concern for the consumer market.
Benchmarks have there place, but, making blanket statements like that is ridiculous
that is why big data centers don't use a "faster" 10W tablets, and news flash, neither does Apple itself
In the server world, 24 months between parts is competitive.
The Epyc lineup has tons of different configs with all kinds of CCD arrangements, from configs with only a single core enabled per CCD with 16 of those, up to all cores for their top end config, with lots of different clock targets, so they could re-use most of the dies.
All info I could find leads be to believe that the majority of the CCDs produced goes into the Epyc lineup:www.3dcenter.org/news/die-marktanteile-fuer-x86-prozessoren-im-ersten-quartal-2025
So, for every 3 sold client CPUs, 2 of those are mobile and 1 is for desktop.
And if for every sold Epyc we have 8 client CPUs sold, but taking into account the above mobile/desktop distinction, we could assume that for every Epyc CPU that's sold, we have 2~3 Ryzen CPUs being sold.
But Ryzen tops out at 2 CCDs, whereas Epyc goes all the way up to 16 CCDs, I think it'd be safe to assume that at worst the CCD allocation is a 50/50 split between Epyc and Ryzen, at best it's 70/30 in favor of Epyc.
newsletter.semianalysis.com/p/tsmcs-heroic-assumption-low-utilization
As for Globalfoundries, they have not upgrade / converted their NY fab from 12nm.You keep saying that Apple is far ahead in performance, but without any evidence.
And Global Foundries is almost certainly in the same spot. Their old, cheap 12nm node is being used for other stuff now.