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

[LittleBro]

It seems that AMD isn't so strict with that, and some motherboards have those 4 lanes shared with M.2 (MSI X870 Tomahawk at least).

AMD was strict about X870(E) having USB4 support. ASMedia USB4 controller (ASM4242) consumes four PCIe Gen 4.0/5.0 lanes (two per USB4 port). There was really no other choice for mobo makers but to implement it CPU-wise, since it could not be possible chipset-wise. So AMD was not so strict, but there was no other choice.

Other points you've listed would be welcome but aren't really related to socket change. AMD could make each of them better or worse with either AM5 or AM6.

Agree, but I think it's vital for AMD to finally introduce more cores per CCD with AM5 and not wait for AM6. There's one thing I forgot to mention and yes it's not socket related, but AMD should really force mobo makers to lower mobo prices. $350 for a mid-range AM5 X870 mobo is really extreme.

Also, bifurcation on CPU-bound lanes. I was expecting to see more flexibility in Zen 5 ... then they announced that the IOD would be reused. Two-lane PCIe slots and M.2 sockets would be usable for many purposes if running at Gen 5 speeds, or even at Gen 4 speeds.

I was thinking about that, too. PEG usually lanes support bifurcation. It would be nice to have 3 M.2 drives connected to CPU using 8 lanes in total (1x4 + 2x2).

AM6 will be a great step up from high-end AM4, with a big +-50% perf uptick, and two more major architecture upgrades to look forward to, as well as decent DDR5 support, quad channel if we're really lucky.

AM6 will be what AM5 was hyped to be.

They say AM6 will come with DDR6.

 

[_roman_]

If AMD really wants for AM5 to live till the end of 2027, with release of Zen 6 they really should:
- make a proper IOD, add native support for CUDIMMS and higher clocks, decrease DDR5 latencies, significantly improve memory controller bandwidth;
- upgrade CPU <> chipset interconnection to PCIe Gen 5 x4;
- make USB 4.0 support chipset-bound, don't hang it onto CPU lanes and stop crippling CPU-bound PCIe lanes;
- add at least 2 more cores per CCD while retaining similar CCD TDP to 8 cores per CCD;
- bring back PCIe ports, since they are versatile, specify for motherboard makers to have more than just two expansion slots (1x PCIe 5.0 x16 + 2x PCIe 4.0 x4 + 1x PCIe 4.0 x 4 is ideal);
- lower native SATA ports, 2 are enough, I'd even prefer motherboards with no SATA support at all (PCIe ports can be converted into SATA ports with adaptor).

EDIT:
- make sure the mainstream mobos for Zen 6 are cheaper!

Maybe you should buy threadripper plattform, especially when you want more PCIE slots. Regardless in what shape. AMD needs to make a big difference between consumer cheap entry hardware and threadripper plattform. Or a second remote box over the network.

Please add your corrections.
I think the cpu socket pins definitions determine what can be "upgraded" or "changed".

People are already buying x870 mainboards although maybe better X670 mainboards are available.

--

I think ASMEDIA designed and makes the chipsets for the amd mainboards, e.g 2 pieces for x670 mainboards.
The usb 4 chip is most often also from ASMEDIA.
I think AMD does not have the knowledge or will to have usb4 in the chip. That will just drive the price up for any processor.

 

[Nhonho]

AMD MUST to abandon the old idea of putting the memory controller (IMC) on the IO die and start putting it on the same die as the x86 cores to the latency of accessing the main RAM be the lowest possible.

Example:

 

[R0H1T]

They say AM6 will come with DDR6.

Who's they?

AMD MUST to abandon the old idea of putting the memory controller (IMC) on the IO die

Old really? Did you forget who had the first IMC probably ever?

 

[Dr. Dro]

I swear, their timings of these announcements is whacked. Right on the heels of the current release... hello?

Despite strong showing for Zen 5 3D in gaming, the architecture as a whole was not a very successful project in the consumer-side. The standard non-3D Zen 5 chips sold poorly in comparison to its predecessors. This is somewhat of a way to save face, saying that a better product is currently on-track for scheduled release. Having a sizable lead in gaming and consumer applications over Intel with the 9800X3D places them in a good position - chips are still a few years away and whoever would skip this, allegedly due to FOMO, are people who don't have money to afford the platform to begin with.

 

[Craptacular]

Who's they?

Anyone who has made this observation:
AM2 chipset supported DDR2
AM3 chipset supported both DDR2 and DDR3
AM4 chipset supported DDR4
AM5 chipset supported DDR5

I trust you can see the pattern.

 

[windwhirl]

AM6 will be a great step up from high-end AM4, with a big +-50% perf uptick, and two more major architecture upgrades to look forward to, as well as decent DDR5 support, quad channel if we're really lucky.

Assuming AM6 won't be DDR6 exclusive. As for the quad channel bit, I'm thinking it will be unlikely, but we'll have to wait to see what happens with that. Though, DDR6 seems to come with four channels per module... so I guess you'll get your wish, just not in the way you expected

Who's they?

Look at AMD's sockets, man

AM2 = DDR2
AM3 = DDR3
AM4 = DDR4
AM5 = DDR5

AMD has always changed the socket when a new RAM standard came out.

As for DDR6, the spec is expected to be finalized either this year or the next, and products usually take a year and half or two years to start showing in the market in large enough quantities. So, 2027 for DDR6 is within expectations and would make sense if AMD decided to launch Zen6 on AM5, since there might not be enough of a DDR6 market at the time to try and sell Zen6 exclusively for an hypothetical AM6 socket

 

[R0H1T]

If AMD is going to maintain compatibility with the old(er) chips then either they have to do dual memory support on zen6 or it will still be DDR5 ~ my bet's on the latter given DDR4 boards/chips are still selling well to this day!

I trust you can see the pattern.

There's no pattern there, it just so happens that AMD needed to keep this going because they're the underdogs. Also who said anything about AM6

 

[freeagent]

Its probably their way of saying, go on, buy that board for 950 beaver bucks.. plus tax.. it will last you a few years..

Look, I wasn't kidding lol.. I contemplated paying the 315 a month haha..

ASUS ROG Crosshair X870E Hero AMD X870E AM5 ATX Motherboard, Advanced AI PC Ready, 18+2+2 Power Stages, DDR5, PCIe® 5.0, 5X M.2, Wi-Fi 7, USB4®, AI Overclocking, Core Flex, PCIe Slot Q-Release Slim : Amazon.ca: Electronics

 

[LittleBro]

Maybe you should buy threadripper plattform, especially when you want more PCIE slots. Regardless in what shape. AMD needs to make a big difference between consumer cheap entry hardware and threadripper plattform. Or a second remote box over the network.

Please add your corrections.
I think the cpu socket pins definitions determine what can be "upgraded" or "changed".

It's not about moving to Threadripper or needing more lanes. I remember x400 or x500 chipset (x = X or B) series motherboards had very good options in terms of PCIe ports. It became worse with x600 series and now with x800 it's terrible. Not everyone uses integrated LAN or audio solution. Now there are plenty 800 series boards that you can't fit those addin cards into.

I am happy owner of ASUS B550-E Gaming, it has a very well done PCIe ports layout.

I am perfectly fine with 24 PCIe Gen 5.0 lanes (16/8+8 for PEG) and 2x4 for NVMe drives. One fast drive for the OS, second for gaming.
Third M.2 in chipset-bound M.2 slot for data. I would not change socket, only upgrade CPU <> Chipset connection to Gen 5.0 x4, which does not need socket to be touched. That would allow for faster M.2 chipset-wise. It was already discussed here on TPU that Zen 5 already has PCIe Gen 5.0 support in terms of connectivity with chipset.

 

[qcmadness]

Despite strong showing for Zen 5 3D in gaming, the architecture as a whole was not a very successful project in the consumer-side. The standard non-3D Zen 5 chips sold poorly in comparison to its predecessors. This is somewhat of a way to save face, saying that a better product is currently on-track for scheduled release. Having a sizable lead in gaming and consumer applications over Intel with the 9800X3D places them in a good position - chips are still a few years away and whoever would skip this, allegedly due to FOMO, are people who don't have money to afford the platform to begin with.

It is a failed architecture that it beats Apple M3 in FP performance.

 

[kapone32]

How is this leak different than anything has done since AM4. People are so quick to forget that in 2024 we got brand new CPUs for AM4. AM5 will be no different. I will promise that the next boards for AM5 should correct the issue with X870E. If you don't want PCIe pay more and get a Mini ITX board or save money and get a Micro ATX. To pay what they want is ludicrous. Everything I have said about MBs was my position before the 9800X3D launch. Anything AMD X3D is way overpriced and sold out. I was blown away when I checked the price of the 7900X3D and it was $999 and sold out on Amazon and Newegg. At least Canada Computers only raised the price by $100 to $799. The 7800X3d is $20 less. We live in a strange time.

It is a failed architecture that it beats Apple M3 in FP performance.

It is not a failed process but that is entirely narrative based. When X3D and Compelling APUs did not exist. Run of the mill CPUs were everything and as such it would be Cyrex vs Intel vs AMD or even IBM if you go back far enough. When Ryzen launched (I have siad this before) it was no different going from 1700X to 2600. Including the bump in Memory performance. Now you have APUs compelling enough that so many companies are making handhelds and Steam is making a real alternative to Windows plus it pushed the price of the 8700G out of HTPC territory. Then X3D became a thing and that feature that gave Intel the crown for having better single threaded performance became it's own animal. I remember people saying that X3D was not a big deal. F me X3D has been so successful that some of the staunchest Intel/Nvida fans on TPU now own X3D chips. TPU and other media sites need to change the way they review chips with this in mind. The future is Single Brand PCs though anyway. If Physx was anything to go by when Nvidia launches their CPU they will probably insert code to make it incompatible with AMD CPUs lol. The CPU Wars have been epic in this generation.

AMD MUST to abandon the old idea of putting the memory controller (IMC) on the IO die and start putting it on the same die as the x86 cores to the latency of accessing the main RAM be the lowest possible.

Example:

Wait, are you saying you know more than AMD engineers? What is the problem? That the regular chips can't clock as high as the APUs?

 

[Dr. Dro]

It is a failed architecture that it beats Apple M3 in FP performance.

I'm not saying it's not good, quite the contrary. But the reviews were clear, unless AVX-512 happens to be involved, the gains over Zen 4 are nominal at best. There is also an architecture-wide regression in power consumption with the 9800X3D using as much as 80% more power than the 7800X3D, so it wasn't entirely a march forward. And it so happened to launch at higher price points as well, so Zen 4 (and in some cases Zen 3) remained attractive and arguably more desirable.

 

[Craptacular]

There's no pattern there, it just so happens that AMD needed to keep this going because they're the underdogs. Also who said anything about AM6

Why isn't the below a pattern?

AM2 = DDR2
AM3 = DDR3
AM4 = DDR4
AM5 = DDR5

 

[Wirko]

I hope the socket AM5 has enough unused contact pads.

It doesn't. The map is here. There are no more than 10 unused pads (light grey, marked RSVD) and even those don't form a single group, they're scattered around.

 

[windwhirl]

Also who said anything about AM6

 

[TumbleGeorge]

It doesn't. The map is here. There are no more than 10 unused pads (light grey, marked RSVD) and even those don't form a single group, they're scattered around.

I don't know what the white squares are without mark. But I see some over-secured, too many of the same function. Perhaps some of them can be redirected to a different function?

 

[Frick]

AM6 will be a great step up from high-end AM4, with a big +-50% perf uptick, and two more major architecture upgrades to look forward to, as well as decent DDR5 support, quad channel if we're really lucky.

AM6 will be what AM5 was hyped to be.

For reference we're currently in the Zen2? stage if we're comparing (maybe even Zen+). Meaning you should compare the 9900X to the 3900X at best, maybe even the 2900X.

 

[AnotherReader]

I'm not saying it's not good, quite the contrary. But the reviews were clear, unless AVX-512 happens to be involved, the gains over Zen 4 are nominal at best. There is also an architecture-wide regression in power consumption with the 9800X3D using as much as 80% more power than the 7800X3D, so it wasn't entirely a march forward. And it so happened to launch at higher price points as well, so Zen 4 (and in some cases Zen 3) remained attractive and arguably more desirable.

There's no regression in power consumption at similar clocks. It also improves performance in regular integer code.

 

[R0H1T]

Why isn't the below a pattern?

AM2 = DDR2
AM3 = DDR3
AM4 = DDR4
AM5 = DDR5

Ok I get it what you're saying, I thought you initially said zen6 was gonna be on AM6

View attachment 371286

It's still speculation as of now, so I won't put too much weight into it. DDR5 will last for a long time!

 

[Dr. Dro]

There's no regression in power consumption at similar clocks. It also improves performance in regular integer code.

I mean, that's the same argument we've seen a while ago "Raptor Lake actually isn't that power inefficient, you can lower the frequencies to reduce power consumption by a ton"; except that both stock and to get all the juice it'll squeeze, you have to make these concessions. Stock for stock IMO counts arguably more than these technicalities in the context of a review and the "standard settings" vast majority of buyers stick to, and they did go ham there a little.

 

[qlum]

No zen 6, in 2025, doesn't mean there won't be anything new. As it stands IO die currently bottlenecking the core, so a refresh with a new io die, and some other improvements could very well be a thing next year.

Plug in an npu, and you got something that makes investors happy.

 

[AnotherReader]

I mean, that's the same argument we've seen a while ago "Raptor Lake actually isn't that power inefficient, you can lower the frequencies to reduce power consumption by a ton"; except that both stock and to get all the juice it'll squeeze, you have to make these concessions. Stock for stock IMO counts arguably more than these technicalities in the context of a review and the "standard settings" vast majority of buyers stick to, and they did go ham there a little.

I agree that stock settings are what matter. I was just pointing out that the architecture is actually more efficient than Zen 4, but probably stung by the Zen 5% idiocy, AMD chose to clock these high.

 

[Vayra86]

That doesn't sound right, didn't AMD claim (announce?) shorter gaps between major launches some time back so the next one is ~2.5 years

Maybe they want to 'focus on midrange' first hm?

I think they're not in a hurry seeing where Intel is going.

I mean, that's the same argument we've seen a while ago "Raptor Lake actually isn't that power inefficient, you can lower the frequencies to reduce power consumption by a ton"; except that both stock and to get all the juice it'll squeeze, you have to make these concessions. Stock for stock IMO counts arguably more than these technicalities in the context of a review and the "standard settings" vast majority of buyers stick to, and they did go ham there a little.

I agree, stock is stock, if AMD wants to place the product as it is, that's what it is. Its the same story on Intel. 'But you can undervolt' ... sure. You can. Until you can't. And you never paid for the performance of last gen clock for clock either; you paid for a 9800X3D with the performance it says it has at the power it uses.

 

[john_]

So, 2027 for DDR6 is within expectations and would make sense if AMD decided to launch Zen6 on AM5, since there might not be enough of a DDR6 market at the time to try and sell Zen6 exclusively for an hypothetical AM6 socket

AMD was usually waiting for Intel to popularize the new memory standard before jumping in. They where always 1-2 years behind. I think AMD will come out with a DDR6 platform AFTER Intel. Assuming Intel exists in 2027 as an x86 CPU designer.