AMD "Zen 6" to Retain Socket AM5 for Desktops, 2026-27 Product Launches

[Chrispy_]

They say AM6 will come with DDR6.

There's a lot of headroom with DDR5 for a while.

With the memory controllers and fabric providing best performance at 1:1 on Zen5, we're at the point where maybe Zen6 will need DDR5-6400 or 6800 at most.

Essentiallly AMD doesn't need and can't really use the full bandwidth of the fastest DDR5 right now because AMD is latency-limited, not bandwidth limited - so the move to DDR6 would actually hurt AMD rather than hinder it. It's why you should buy DDR5-6000 CL28 and tune it for tighter timings rather than faster frequencies, because all Zen5 wants is lower latencies.

 

[LittleBro]

I don't really know if Ryzen will profit as much from faster memory. Isn't it a good thing AMD isn't ad memory sensitive as Intel?

X3D technology clearly shows that Zen architecture is memory starved. On non-X3D CPUs you can easily get +10% perf. just by tightening memory timings (lowering latency).

AMD has shitty IMC, which cannot really run with DDR5 in 1:1 mode as with DDR4. There is a latency penalty. Hopefully that finally changes with Zen 6

I fully agree. They could use 4 additional lanes from the CPU for Gen4x8 or Gen5x8, but that would limit available lanes even more-

This is not the way. Using another 4 CPU lanes will require chipset to provide 4 lanes. That would leave only very few lanes for chipset.
Upgrading PCIe Gen 4.0 x4 CPU<>Chipset interconnection to PCIe Gen 5.0 x4 doubles the bandwidth and no additional lanes are required.

No, put USB4 into the IOD. It already is in Phoenix/Hawk Point and Strix Point. Sadly, only one B650E mainboard from Gigabyte supports using USB4 from CPU.

I don't think you just add 2 Cores, you probably only can double core count to 16.

And why not? Ryzen Z2 Extreme has 3+5 cores. You can also disable 6 defective cores on a chip of 16, that will get you 10-cores.

You mean offer more PCIe lanes so motherboard makers are able to offe more PCIe-slots. AMD can't force them to put those lanes into PCIe-slots. That being said, sadly, M.2 and gigantic GPU-coolers are killing space for PCIe-slots.

No. Just bring back good old PCIe lanes distribution into 3-5 PCIe slots. No additional PCIe lanes needed.

With the memory controllers and fabric providing best performance at 1:1 on Zen5, we're at the point where maybe Zen6 will need DDR5-6400 or 6800 at most.

Best performance would be 1:1:1: (UCLK:FCLK:MCLK) as with Zen 3. IF and IMC are slow.

Essentiallly AMD doesn't need and can't really use the full bandwidth of the fastest DDR5 right now because AMD is latency-limited, not bandwidth limited - so the move to DDR6 would actually hurt AMD rather than hinder it. It's why you should buy DDR5-6000 CL28 and tune it for tighter timings rather than faster frequencies, because all Zen5 wants is lower latencies.

Agree on the fact that AMD can't fully utilize DDR5, there's absolutely no need for DRR6 yet. AM6 will most probably come only with DDR6. Increasing memory frequency not only increases bandwith, but also decreases memory latency. Sometimes you can raise frequency and timings can remain (almost) the same. I was unable to tighten my B-Die DDR4s more with <1.50V, so I tried to go for higher frequency instead, needed just to very slightly raise the VDIMM to make it stable in HCI MemTestPro and with same timings and additional frequency bump the latency improved by roughly 3 ns. There are memory chips which handle better higher frequency and then there are those which can't run that fast, but can be tightened more. A chip may require to raise VDIMM much more to achieve tighter timings than to achieve higher frequency.

As other have already pointed out, it'd really help Zen to move IMC out of IOD, much closer to cores.

 

[Garrus]

Zen5 turned out to be awful. A complete flop and had no reason to release. Intel Core 200 is even worse, worse than their old stuff.

Meanwhile the same CPU core in the 9800X3D is excellent. 20+ percent better in demanding games, lower temperatures, higher clock speeds, even much improved productivity performance.

Shows that the I/O die or memory system for the non 3D parts needs major improvements.

9800X3D is basically the only CPU you should buy, imo. Release a bunch of X3D parts and stop selling the rest.

I'd like to see 9800G with 16 GPU cores release as well. 9600G with 12 GPU cores. That's it.

 

[wheresmycar]

What an interesting upgrade path for those early AM5 adopters... especially the non-X3D gamers with a 7600/7600X/7700X. IF/When the time calls for it, something like a 10800X3D would be a tasty booster. AMD nailed it with AM4 and it seems AM5 will repeat. Who knows maybe some enriched Zen 6 'refreshes' further down the 2027/+ trail.

For me, while I appreciate performance improvements in CPUs, what stands out most to me is the advancement in GPUs and the big FAT question mark around their pricing.

 

[LittleBro]

Zen5 turned out to be awful. A complete flop and had no reason to release. Intel Core 200 is even worse, worse than their old stuff.

It's not like that. On Windows, performance uplift is low. On Linux-based systems, we're talking about 12-15% on average. There is a huge performance improvement for AVX512 workloads, which are not supported by anything newer than Intel's Alder Lake. Then, there is Windows 11 and related performance issues with both Zen 4/5 and Intel Arrow Lake.

Optimierte Sprungvorhersage: Zen 5 mit Windows-Update KB5041587 getestet - Hardwareluxx

Optimierte Sprungvorhersage: Zen 5 mit Windows 11 24H2 getestet.

www.hardwareluxx.de

What is worse with Arrow Lake launch is it's launch along with problems like BSODs, instabilities. Intel said some time ago that they'd prefer delivering on time rather than delivering expected results. I'd say both Intel and AMD now cook their hardware much faster than before which leads to release of unpolished products.

Ryzen 5 9600X is a good processor, currently costs around 240 € incl. VAT. I do recall how I paid 310 € for mine 5600X few months after Zen 3 release. Those who aim for 9600X can't really complain now, it's a solid deal. Ryzen 9800X3D is best gaming processor thanks do 3D cache. Same can be expected for upcoming 9900X3D and 9950X3D. I'd say 9600X3D will be best gaming processor when we take into account price (7800X3D performance at around 300 €). Arrow Lake will get better with patches. Windows 11 is a piece of shit software.

What is much much worse though is how it looks on the market now: https://x.com/TechEpiphanyYT/status/1855904874883338582
It seems Intel is currently experiencing some harsh times. They probably won't beat Q4 expectations due to poor sales. They really need to get their together and start kicking again. Their Lunar Lake SKU variety is absolute overkill. Sometimes less is more.

 

[Chrispy_]

AMD nailed it with AM4 and it seems AM5 will repeat.

They didn't "nail it" - they simply don't change socket every thirty seconds, rather they change socket when there's a major new shift in other hardware compatibility (ie, memory type, or the industry move from AGP to PCI express)

I've yet to see any credible evidence for why Intel needed to change the socket so many goddamn times. Take the Core series from 2008 and look at the pin count/architecture:

• 1st Gen Nehalem: 1156 pins, DDR3, PCIe
• 2nd Gen Sandy bridge 1155 pins, DDR3, PCIe
• 4th Gen Haswell 1150 pins, DDR3, PCIe
• 6th Gen Skylake 1151 pins, DDR3 (and optional DDR4), PCIe

That's four new platforms in just six years, all with about the same number of pins and basically identical connectivity. 1156 and 1150 were particularly awful - not even getting a single upgrade generation was ridiculously harsh on those buyers. Was there any technical reason or killer feature to explain why newer processors couldn't have gone into an older socket? No. It was a spiteful marketing decision of forced obsolescence to sell motherboards unnecessarily.

Let's continue:

• 12th Gen Alder Lake, 1700 pins, DDR5, PCIe
• S2 (15th Gen?) Arrow Lake, 1851 pins, DDR5, PCIe

So they increased the pin count from 9th/10th/"11th" gen by 500, mostly for more power delivery from what I can tell, with some additional redundancy, then as soon as a new microarchitecture comes out with lower power consumption and the exact same DDR5/PCIe width, Intel decide we need even more pins? What the hell, Intel? Your own pinouts show that there are more than 151 spare, unused pins. Perhaps LGA1851
is the start of a platform longevity push and all those spare pins are there for new processors that will be compatible in four generations from now - but given all the evidence so far, my money is on Intel ditching 1851 long before any processors that actually use 1851 pins turn up on the scene!

Ignoring all that unnecessary socket switching, Intel failed to keep mounting hardware and mounting holes consistent through most of those platform changes, so not only are you throwing out the motherboard, you're also forced to buy a new cooler, or at least hope your cooler manufacturer provides updated hardware.

 

[A Computer Guy]

They didn't "nail it" - they simply don't change socket every thirty seconds, rather they change socket when there's a major new shift in other hardware compatibility (ie, memory type, or the industry move from AGP to PCI express)

I've yet to see any credible evidence for why Intel needed to change the socket so many goddamn times. Take the Core series from 2008 and look at the pin count/architecture:

• 1st Gen Nehalem: 1156 pins, DDR3, PCIe
• 2nd Gen Sandy bridge 1155 pins, DDR3, PCIe
• 4th Gen Haswell 1150 pins, DDR3, PCIe
• 6th Gen Skylake 1151 pins, DDR3 (and optional DDR4), PCIe

That's four new platforms in just six years, all with about the same number of pins and basically identical connectivity. 1156 and 1150 were particularly awful - not even getting a single upgrade generation was ridiculously harsh on those buyers. Was there any technical reason or killer feature to explain why newer processors couldn't have gone into an older socket? No. It was a spiteful marketing decision of forced obsolescence to sell motherboards unnecessarily.

Let's continue:

• 12th Gen Alder Lake, 1700 pins, DDR5, PCIe
• S2 (15th Gen?) Arrow Lake, 1851 pins, DDR5, PCIe

So they increased the pin count from 9th/10th/"11th" gen by 500, mostly for more power delivery from what I can tell, with some additional redundancy, then as soon as a new microarchitecture comes out with lower power consumption and the exact same DDR5/PCIe width, Intel decide we need even more pins? What the hell, Intel? Your own pinouts show that there are more than 151 spare, unused pins. Perhaps LGA1851
is the start of a platform longevity push and all those spare pins are there for new processors that will be compatible in four generations from now - but given all the evidence so far, my money is on Intel ditching 1851 long before any processors that actually use 1851 pins turn up on the scene!

Ignoring all that unnecessary socket switching, Intel failed to keep mounting hardware and mounting holes consistent through most of those platform changes, so not only are you throwing out the motherboard, you're also forced to buy a new cooler, or at least hope your cooler manufacturer provides updated hardware.

On the flip side I suspect when most people upgrade their PC's it's a long enough period of time a socket change would have been needed anyway. For example 3 to 5 years between upgrades often ending up in the replacement of the entire PC. Enthusiasts upgrade at a faster pace and in todays world end up often being the conscripted beta testers.

In my opinion AMD's long term socket strategy wouldn't have been worth it if it weren't for the ability to jump to 16-core or X3D chips.

 

[Chrispy_]

On the flip side I suspect when most people upgrade their PC's it's a long enough period of time a socket change would have been needed anyway. For example 3 to 5 years between upgrades often ending up in the replacement of the entire PC. Enthusiasts upgrade at a faster pace and in todays world end up often being the conscripted beta testers.

If someone always buy new, current-gen hardware and never does any upgrades until you buy all-new again, the socket discussion is completely irrelevant to that person. In that aspect, what you say is entirely correct.

Take AM4 though; Even before AM5 launched, it had been around long enough that there was an enormous selection of motherboards for all sorts of different budgets and form factors. You could dump anything from an ultra cheap used A6 APU into it, or you could slap a brand new 5950X into it. There was more CPU choice at the end of AM4's reign than the cumulative total of five entire socket changes from Intel, and that huge, multi-generational range had used options that simply didn't exist in the market for intel because nothing compatible was old enough to even appear on the used market.

The problem extends beyond even just the used market. Brand new "last-gen" stock on clearance was hampered by incompatibility and the value of those things vanished as stocks of discontinued motherboards with the correct sockets dry up, nullifying the appeal of a last-gen processor on sale at deep discounts.

 

[Count von Schwalbe]

• 6th Gen Skylake 1151 pins, DDR3 (and optional DDR4), PCIe

You forgot the two incompatible versions of 1151, that there was no technical reason to be incompatible. After all, you could tape off a couple of pins on newer CPUs to put them in older boards.

 

[kapone32]

You forgot the two incompatible versions of 1151, that there was no technical reason to be incompatible. After all, you could tape off a couple of pins on newer CPUs to put them in older boards.

There is no comparison between AMD and Intel in regards to Socket support.

 

[3valatzy]

AM6 will be a great step up from high-end AM4

Hopefully, sometime around year 2035 with DDR6.

 

[wheresmycar]

They didn't "nail it" - they simply don't change socket every thirty seconds, rather they change socket when there's a major new shift in other hardware compatibility (ie, memory type, or the industry move from AGP to PCI express)

I've yet to see any credible evidence for why Intel needed to change the socket so many goddamn times. Take the Core series from 2008 and look at the pin count/architecture:

• 1st Gen Nehalem: 1156 pins, DDR3, PCIe
• 2nd Gen Sandy bridge 1155 pins, DDR3, PCIe
• 4th Gen Haswell 1150 pins, DDR3, PCIe
• 6th Gen Skylake 1151 pins, DDR3 (and optional DDR4), PCIe

That's four new platforms in just six years, all with about the same number of pins and basically identical connectivity. 1156 and 1150 were particularly awful - not even getting a single upgrade generation was ridiculously harsh on those buyers. Was there any technical reason or killer feature to explain why newer processors couldn't have gone into an older socket? No. It was a spiteful marketing decision of forced obsolescence to sell motherboards unnecessarily.

Let's continue:

• 12th Gen Alder Lake, 1700 pins, DDR5, PCIe
• S2 (15th Gen?) Arrow Lake, 1851 pins, DDR5, PCIe

So they increased the pin count from 9th/10th/"11th" gen by 500, mostly for more power delivery from what I can tell, with some additional redundancy, then as soon as a new microarchitecture comes out with lower power consumption and the exact same DDR5/PCIe width, Intel decide we need even more pins? What the hell, Intel? Your own pinouts show that there are more than 151 spare, unused pins. Perhaps LGA1851
is the start of a platform longevity push and all those spare pins are there for new processors that will be compatible in four generations from now - but given all the evidence so far, my money is on Intel ditching 1851 long before any processors that actually use 1851 pins turn up on the scene!

Ignoring all that unnecessary socket switching, Intel failed to keep mounting hardware and mounting holes consistent through most of those platform changes, so not only are you throwing out the motherboard, you're also forced to buy a new cooler, or at least hope your cooler manufacturer provides updated hardware.

That's the very definition of AMD "nailing it" from a consumer perspective. The context here was how early adopters of the platform benefited from AM4’s long-term support, especially compared to the constant socket swaps we saw with Intel. I missed out, but the cost effective ecstasy of jumping from Ryzen 1000-series (or even 2000) to a mouth-watering 5800X3D/5950X is just insane!

I agree, Intels socket strategy and short of appealing incremental performance gains which came at a higher cost was frustrating enough to discourage feeding the upgrade rage. On a positive note, it did help tame the upgrade-itch for a bit.

 

[Chrispy_]

You forgot the two incompatible versions of 1151, that there was no technical reason to be incompatible. After all, you could tape off a couple of pins on newer CPUs to put them in older boards.

Lol, I'd forgotten about that nonsense - mostly because when Intel launched Coffee Lake on their silly "new" S1151 they were already well behind AMD and anyone actually wanting more than 4 cores had already been running 8-core Ryzens for 8 months at that point.

IIRC Coffee lake sat in a no-man's land. On paper it was a really really good CPU but in reality the much cheaper 6th Gen and 7th Gen i5 quad cores were close to or identical in gaming performance, and Ryzens were absolutely murdering Intel in multithreaded applications because, unsurprisingly, 8 cores are more than 6 cores. By the time any significant percentage of games were starting to see the benefits of more than four cores, Zen2 was already the CPU to get.

That's the very definition of AMD "nailing it" from a consumer perspective. The context here was how early adopters of the platform benefited from AM4’s long-term support, especially compared to the constant socket swaps we saw with Intel. I missed out, but the cost effective ecstasy of jumping from Ryzen 1000-series (or even 2000) to a mouth-watering 5800X3D/5950X is just insane!

Well yeah, if your definition of "nailing it" is "don't be assholes" then yeah, AMD sure did nail it.

 

[RandallFlagg]

Maybe Zen 6 will provide more improvement than we've seen recently, but I doubt it.

The last 3 years shows performance gains with CPUs slowing down a lot. We're in a place where a 2 year old 13900K is a mere 6% slower than a brand new release 9950X in overall application performance. Sure you can cherry pick something, but overall the performance gains are undetectable by any normal user.

Unless you are a hard core gamer with top of the line GPU, which many here seem to be, there's virtually no reason to get amped up about these new sockets and CPUs.

 

[Dawora]

Nice. Time to buy me a AM5 board i think. Good thing i stuck to the 12700k as it's time to catch another bus

And whitout strong GPU u dont see difference.. even more so if u gaming 1440p

Maybe Zen 6 will provide more improvement than we've seen recently, but I doubt it.

The last 3 years shows performance gains with CPUs slowing down a lot. We're in a place where a 2 year old 13900K is a mere 6% slower than a brand new release 9950X in overall application performance. Sure you can cherry pick something, but overall the performance gains are undetectable by any normal user.

Unless you are a hard core gamer with top of the line GPU, which many here seem to be, there's virtually no reason to get amped up about these new sockets and CPUs.

And thats whit 4090+1080p
Cpu not matter much realy, its more about GPU

 

[LittleBro]

AMD should definitely visit TPU to talk with us on how Zen 6 should be.

 

[TumbleGeorge]

AMD should definitely visit TPU to talk with us on how Zen 6 should be.

To this end, let's repeat the necessary actions. Now that AMD has expanded the front end of the architecture, it needs to remove the rest of the bottlenecks. The slow Infinity factory, the outdated RAM controller and the generally outdated cIOd. Of course, all these actions will also need additional tweaks on the front-end, which is currently tailored for the old version of cIOd and the old and slow Infinity factory. Hopefully, however, the AM5 socket will allow everything new to work well and is not the final bottleneck that is insurmountable.
Well, the problem remains how these changes can actually be made. Are they even possible, are they cost-effective and competitive, and...can they deliver sufficient growth in real-world applications. Because this will again require improved drivers and...improvement in applications to take full advantage on the new architecture.

 

[Zach_01]

I want to believe that AMD will improve the IOD on most aspects and that its going to be on AM5. Maybe on next Zen6 or even a Zen5+.

 

[kapone32]

Has AMD ever given us a 2% improvement over generations since Ryzen? Are their CPUs slow? Is 9000 faster than 7000 and 5000? 10% is actually fine when you step back. The Memory controller is only a good thing if it can increase in speed. I mean if 64MB of Cache can have such an impact on Gaming a faster Memory controller would be great. Something like the controller on the APUs.

 

[Zach_01]

Something like…

FCLK: 2666MHz
UCLK: 4000MHz
So it could do 1:1 with 8000MT/s
MCLK: 4000MHz

…would be great

As I said before this would increase package power substantially, especially the FCLK speed and that’s why it needs to be on a better fab node than current ones Zen5 has. CCD and IOD.

 

[Chrispy_]

Something like…

FCLK: 2666MHz
UCLK: 4000MHz
So it could do 1:1 with 8000MT/s
MCLK: 4000MHz

…would be great

As I said before this would increase package power substantially, especially the FCLK speed and that’s why it needs to be on a better fab node than current ones Zen5 has. CCD and IOD.

I'm no expert in CPU architecture, but is there any point to non-integer ratios of FLCK to UCLK? The data is either there on a cycle or it's not. Running the FCLK at 2666 would, as far as I understand it, not achieve anything other than increase power draw over an FCLK of 2000.

If I'm wrong I wouldn't mind an ELI5 as to how non-integer ratios can provide a benefit.

 

[Zach_01]

It just a thought.
I just kept the 2:3:3 ratio that most do now and add 33% higher frequency.

2000:3000:3000 (6000MT/s)
2667:4000:4000 (8000MT/s)

Of course the best of the best is 1:1:1 but in order to keep that you have to increase FCLK too much for high frequency DRAM or use low frequency DRAM
Another option is keep 1:2 the UCLK:MCLK which is not the best option either as UCLK is too(?) low.

For example
FCLK:UCLK:MCLK
1:1:1
3000:3000:3000 (6000MT/s)
2667:2667:2667 (5333MT/s)
2400:2400:2400 (4800MT/s)

1:1:2
2400:2400:4800 (9600MT/s)
2333:2333:4667 (9333MT/s)

Maybe the 2:3:3 is the best middle ground (?).

 

[StimpsonJCat]

Has AMD ever given us a 2% improvement over generations since Ryzen? Are their CPUs slow? Is 9000 faster than 7000 and 5000? 10% is actually fine when you step back. The Memory controller is only a good thing if it can increase in speed. I mean if 64MB of Cache can have such an impact on Gaming a faster Memory controller would be great. Something like the controller on the APUs.

AMD was hyping gains of well over 20%, right up to launch... Then they launched it and Zen 5% was born. Then their marketing team went into damage control...

 

[kapone32]

AMD was hyping gains of well over 20%, right up to launch... Then they launched it and Zen 5% was born. Then their marketing team went into damage control...

Do you mean the 9000? Wasn't it confirmed that Windows was holding back the CPU?

 

[StimpsonJCat]

Do you mean the 9000? Wasn't it confirmed that Windows was holding back the CPU?

Yes, AMD was saying it about the Zen 5 core itself, not a particular SKU. And yes, there was a small percentage loss due to a bug in Windows VBS security. But it was a 1-4% loss, in SOME games. After the patch, Zen 5 remains on average, 5% faster than Zen 5 in gaming. It does a little better in some productivity, but not much.