AMD Zen 6 CPUs confirmed to work on existing AM5 motherboards | Asus and Asrock confirm Zen 6 support, next-gen Ryzen CPUs on track for early 2027
Posted by chrisdh79@reddit | hardware | View on Reddit | 273 comments
vegetable__lasagne@reddit
Is Zen 6 expected to be a big change from Zen 5? Or is it mainly a core count increase?
Swaggerlilyjohnson@reddit
It will probably be the biggest upgrade since OG zen but I also expect it to be the biggest price hike for the same reason.
They are going from 5/4nm to 2nm so it's two node jumps (Skipping 3nm). They are rumored to be significantly upgrading the IO die and going from 8 to 12 cores and significantly increasing frequency and IPC.
Even auxiliary downsides we have been dealing with since the original move to chiplets (Really high idle power) are rumored to be fixed.
Don't expect it to be cheap though. I wouldn't be surprised if AMD kept producing large amounts of Zen 5 just because it will be expensive and the budget segment won't want to pay for 2nm and more cores.
Vb_33@reddit
How'd they fix that
Swaggerlilyjohnson@reddit
A similar approach to how strix halo works. The rumor is they are using a fanout interconnect instead of their traditional infinity fabric interconnect. High yield talks about it in depth if you are interested in how it works https://www.youtube.com/watch?v=maH6KZ0YkXU
These are rumored improvements but it is more concrete then most rumors because they already used the technology on strix halo so it's not just possible but they have already done it on another consumer product.
It already massively helped with strix halo it's unknown if they improved it even further but even if they just use the same design we have already seen it make a massive difference.
You can look up the phawx videos on strix halo to see how efficient it is below 15w (Not Idle but like an entire 40cu IGPU and 16 cores running that low of a wattage. The actual idle is orders of magnitude lower than their older chiplet designs)
Earthborn92@reddit
Zen 6 is N2. It's basically two node jumps.
Quite a big improvement on just the process alone, even if there aren't large IPC gains
BlueSiriusStar@reddit
That's maybe for server for client it may be N3 or N4 to save costs.
Vb_33@reddit
Desktop is N2 just like server will be.
BlueSiriusStar@reddit
But why, though, are the server CCDs the same as the Desktop CCDs?
petuman@reddit
AMD shares/reuses CCDs between desktop and server (so far), yeah.
BlueSiriusStar@reddit
Oh really, I didn't know. Thought server would have better CCD dies to use, which may have more cores counts idk.
Earthborn92@reddit
The whole chiplet approach is based on reusing the core dies between platforms
EloquentPinguin@reddit
Server and Desktop N2, Laptop N3, servers will launch early 2026 with N2B, laptops mid/late 2026 with some N3, and desktops will be late 2026/early 2027 with improved N2
BlueSiriusStar@reddit
CCD could be N2 and IO Die could be N4 this may improve over years if the yields prove good. N2 is expensive doubt AMD would fab all the dies using N2 unless it's for Medusa Point.
EloquentPinguin@reddit
Yeah I wouldn't expect the IOD to be N2. Either a cheap N3 node or a mature N4 node but N2 for the IOD seems like a lot of waste.
CrzyJek@reddit
Zen6 is using a significantly better IO die, bridge dies, and the node is designed for very high clock speeds. You'll see clocks probably between 6-6.6ghz speeds.
ResponsibleJudge3172@reddit
6.6ghz? Hold your horses
Exist50@reddit
What's unreasonable about that number? Just from process gains alone, you'd expect something in the ballpark.
fiah84@reddit
that's what Intel thought when they made the Pentium 4
yes this time it's different but I'll believe it when I see it
Exist50@reddit
I'm not sure what the concern is supposed to be. We saw an 800MHz jump going from Zen 3 to Zen 4, and that was just one node. Even a fully 1GHz increase with Zen 6 wouldn't be absurd. Same design team as Zen 4, btw.
fiah84@reddit
it wouldn't be completely absurd, it could happen, however there are no guarantees. People have spent literally decades trying to ship general purpose CPUs at these frequencies, the ones that got closest recently not only needed a metric shitton of power to get there but also started frying themselves
so if AMD somehow pulls a 6gHz Zen6 out of their hat and ships it, I will be suitably impressed
Exist50@reddit
People not too many years ago claimed 5GHz was an absurd, unobtainable value as well. Now, no one bats an eye. Likewise, there's nothing special about 6GHz.
This is not a Pentium 4 type of situation. It's just expected, generational changes.
fiah84@reddit
you should definitely manage your expectations then
Exist50@reddit
Again, Zen 4 saw even greater gains with a lesser node shrink.
ResponsibleJudge3172@reddit
Zen 4 didn't also make major changes and increase core count by 50%
Exist50@reddit
And? None of the SoC-level changes should influence our expectations from a core perspective.
wintrmt3@reddit
The netburst was a whole new design that sacrificed everything for clock speeds, but it did double it compared to the p3, 20% more clocks hardly compares to that situation.
SIDER250@reddit
If this is to be believed
https://www.reddit.com/r/radeon/comments/1nwr5gz/zen6/
I am not sure then
Toojara@reddit
Seems made up, the L3 for the 990X is blatantly wrong and likely is so for the new cut down parts. I also doubt they'd round IPC like that if they knew what they were doing.
pceimpulsive@reddit
Well IO die is likely to change so yes it should be a decent change. I'd expect better memory subsystem performance,
burd-@reddit
should we see low idle power or still high due to chiplet design
dsoshahine@reddit
Rumour is that Zen 6 is adopting the fanout interconnect from RDNA3/Strix Halo and Strix Halo is amazing with idle performance (about as good as the monolithic mobile parts). Hopeful this will translate well into desktop.
pceimpulsive@reddit
Yeah this is actually a big hope! Looks amazing, 256bit memory interface on desktop CPU :O
petuman@reddit
Surely 256 bit bus isn't coming to desktop? AM5 socket isn't wired for it to start with.
Maybe with AM6 we'll see wider memory bus (as 2DPC is going away anyway)
pceimpulsive@reddit
The new fanout enables that as it has 8 32 bit memory phy on the IODie
Well probably get 4 disabled~
trololololo2137@reddit
strix halo idle power is a disaster https://www.notebookcheck.net/Asus-ROG-Flow-Z13-GZ302EA-Convertible-Review-AMD-s-Strix-Halo-GPU-is-neck-and-neck-with-the-RTX-4070-Laptop.963266.0.html mentions 10W on idle MINIMUM which is like 3-4x higher than what it should be
pceimpulsive@reddit
I imagine no drops in idle power. Pets see though?
no6969el@reddit
I'm excited to see the evolution of the x3d lineup.
pceimpulsive@reddit
Me too!
I'm still on 5800X3D. Am itching to upgrade to zen6 aye...
gnarlysnowleopard@reddit
I'm on the 5800X3D and looking to skip AM5 altogether. There's not a single game I play where I feel like it's leaving performance on the table.
Addendum-Haunting@reddit
Tempted to skip AM5 but with zen 7 confirmed to be on AM5. Zen 7 likely to release around 2028-2029. Long wait for AM6. Only time will tell though, we will see how zen 6 stacks up
gnarlysnowleopard@reddit
where did you hear that it's confirmed Zen 7 will be on AM5?
Addendum-Haunting@reddit
Moore’s law is dead YouTube channel. Essentially had official documentation of it.
I647@reddit
Same. Its truly one of the goats.
pceimpulsive@reddit
I was too, bower it looks like there could be zen7 on am5 as well... We might be waiting till 2029 for AM6 (zen successor).
I'm feeling ddr4 mostly and my x370 boards IO/pcie constraints sorta suck!
BlazingSpaceGhost@reddit
Same I'll be on the AM6 train when it leaves the station but for now there is no need to upgrade. I however was a fool and about the 5800x and not the x3d so I do feel it in some games now but it's not worth the upgrade price yet.
Stingray88@reddit
Same. I can’t justify buying the last gen on AM5 when I could instead buy the first gen of AM6 and hopefully have sooo much more runway for future upgrades. Upgrading to my 5800X3D was a wonderful experience and I want to see that again on the same platform.
theholylancer@reddit
If I had a 5800x3d I'd wait for am6 and either ddr6 or camm based stuff
I highly doubt even with real fast ddr6 it would significantly increase on that front
Zen 6 with new io die but still ddr 5 won't be as good as camm2 or ddr 6 I think
pceimpulsive@reddit
I've been waiting for AM6 too but it looks like maybe zen7 might happen on AM5 as well, and my internal parts are starting to age!
My dram, board, PSU are from 2017...
I can only have 1 nvme SSD, and I cannot use all my PCIE lanes due to motherboard wiring layout.
I want more storage, I want more CPU cap, and this old AM4 X370 can't deliver that!
We will see what zen6 looks like and if it will be the last gen chip for am5 or not...
Depends when ddr6 comes to desktop too.
ammar_sadaoui@reddit
what is wrong with your CPU to upgrade?
pceimpulsive@reddit
General performance.
Mostly my mobo is lacking still on X370.
mycall@reddit
If x3d became standard on all of their chips, that would be fantastic.
Vb_33@reddit
Sounds pointlessly expensive.
mycall@reddit
Why does expensive have to involve your comment for my needs?
gusthenewkid@reddit
Not everybody does tho, hence why it’s a stupid idea.
Plank_With_A_Nail_In@reddit
7% faster for 20% more money, if you are expecting anything else you are daft.
no6969el@reddit
If only I can go a day without someone on Reddit being an asshole.
INITMalcanis@reddit
3D cache helps reduce the performance loss from Zen5's slightly outdated memory subsystem, so it might be a less significant factor for Zen5.
Standard-Potential-6@reddit
Finally AV1 encode
Exist50@reddit
Well the shrink to N2 is quite significant. There're some crazy clock speed rumors I do not believe, so will not repeat, but somewhere in the mid 6GHz range would not be surprising. And some IPC gains (~10%?) on top. That plus the 1.5x core count and reworked IO die should make it quite interesting.
NerdProcrastinating@reddit
Looking at David Huang's SPECint2017 testing perf/GHz data and comparing an M4 Pro 3.04 / GHz against 9800X3D (2.40) and Strix Halo (2.08) shows Apple's M4 Pro currently at \~26.6% to 46% faster per clock than AMD's best.
Zen 6 only getting \~10% IPC gains would be pretty disappointing. If we assume the M5 Pro is around the same \~10%, then Zen 6 will still leave us with the current large perf/clock gap (which will roughly translate to perf/watt).
Plank_With_A_Nail_In@reddit
its going to be 7% gains for 20% more cost, don't fool yourself into thinking you are going to get 30% faster.
Exist50@reddit
You think a 2 node shrink will only net them 7% gains? To say nothing of the core count increase...
Don't be ridiculous.
Silv_St@reddit
Mainly the core count and a big node jump, but if I had to guess Infinity fabric changes and maybe a change to their 8-wide decoder
Beefmytaco@reddit
Better be a change to core count. At this point AMD has perfected the current amount and it's time to move to the next level.
The 10800X better be at least 10c20t.
hackenclaw@reddit
I dont know, I kinda prefer they stick to same core count but add 4 cores into IO die, making it 16+4.
This would allow the CPU to shut down all the chiplet while idling, on windows/OS.
Plank_With_A_Nail_In@reddit
We are talking about a desktop chip here why on earth are you worried about idling?
Vb_33@reddit
Why do we want to mindlessly waste power.
kat0r_oni@reddit
Because I do not like the 40W idle power my CPU uses?
dsoshahine@reddit
Why on earth are you not worried about idling? Personal computers spend most of their time idling or in low-load scenarios. Dropping idle power to near mobile levels would be amazing. This is one of the last areas Intel CPUs still have been superior in vs AMD's chiplet CPUs. Why wouldn't they try to improve this?
hackenclaw@reddit
because the same desktop chip are use on HX laptop.
AMD HX idling is their biggest weakness against Intel completing product. This market isnt a small market to ignore.
Vince789@reddit
Yep, but I believe Strix Halo fixed it's idle power consumption (without adding E cores on the IO+GPU Die) by changing the interconnect to InFO-oS (Integrated Fan-Out on Substrate)
Maybe AMD might still add LPE cores to the IO die, but InFO-oS alone should fix idle power consumption
New_Enthusiasm9053@reddit
Also IO is fabbed on an older node. Don't want to see the scheduling issues caused by that lol. Intels IO E cores already don't ever get scheduled as far as I can tell on my work laptop(2 cores are presumably not enough for windows even at idle).
Vb_33@reddit
It'll be 12c24t. The more curious part will be the 10900X as we know the 10950X will be 24c48t.
Exist50@reddit
Rumors are for a 12c CCD.
RedIndianRobin@reddit
Higher chance of Earth getting struck with an asteroid than AMD giving higher core count CPUs lol. They're the new Intel of the modern era. Why innovate when they can milk consumers dry since their competition sucks now?
Plank_With_A_Nail_In@reddit
Lol displaying ignorance of CPU design and asteroids amazing achievement there dude lol.
based_and_upvoted@reddit
If you count the atmosphere as the earth then we get luke 15000 impacts per year, but if you only count the ones that hit the ground i think the last one was in 2013 or so
Plank_With_A_Nail_In@reddit
Around 5 hit the ground as a solid lump. Around 500 tons of debris hits the ground. "Burning up" is a misnomer most of the dust hits the ground.
9897969594938281@reddit
You need to think about their server offerings, and how a higher core count per chiplet would help
panchovix@reddit
I just hope more PCIe lanes, but if they keep AM5 then I guess not much hope.
FinishIllustrious116@reddit
What is problem to add more PCIe lanes in south bridge(MB chipset)? Socked doesn't affect in this cause.
Plank_With_A_Nail_In@reddit
Its hard to run mutliple GPU's and lots of NVME sockets at the moment. Need at least 4 NVME and would like to run 4 GPU's without needing to buy threadripper or the like.
soggybiscuit93@reddit
Idk if we'll ever see 4 dGPUs (without compromise) on consumer platforms. Its just so much much IO (and additional cost) to support such a niche setup on consumer boards.
Those types of setups are exactly why HEDT / workstation exist.
BlueSiriusStar@reddit
But, increasing the number of PCIE SouthBridge lanes will still use the same DMI PCIe line to the CPU. After the link is saturated our benefits gained would be limited
FinishIllustrious116@reddit
It leaks are true, south bridge to CPU i/o chiplet will be upgraded from PCIe 4.0 x4 to PCIe 5.0 x4.
BlueSiriusStar@reddit
But then you would still need a new motherboard to benefit from the extra lanes, which defeats the purpose of a continuous socket if the motherboard manufacturers can give us a better socket from the start.
Stingray88@reddit
No it doesn’t, because not everyone will need those extra lanes.
BlueSiriusStar@reddit
Really alot of people would benefit from more PCIE NVMe slots maybe. Higher Speed I/O, more 10G ports, Thunderbolt 5, possibilities are endless and people's needs might change.
UsernameAvaylable@reddit
Eh, i would be most people will never use more than 1 (one) nvme slot.
Stingray88@reddit
Some would benefit from all that, but a lot wouldn’t. Personally, with how I use my desktop today, I wouldn’t at all. I’ve got one GPU, two NVMe SSDs, and exactly zero high bandwidth requiring peripherals. The GPU and one SSD are connected directly to the CPU. The 2nd SSD is on the chipset, but it’s not really competing with anything for bandwidth as it is.
And again, I’m not saying my use case covers everyone. Surely some people would benefit from higher bandwidth to the chipset. The point is simply that performance improvements that come to later motherboards on the same CPU socket do not defeat the purpose of a continuous socket. Lots of people who don’t need more bandwidth to their chipset could still benefit from a new CPU generation.
FinishIllustrious116@reddit
Yes, motherboard with next generation chipset may unlock full features. People's which upgrade AMD ZEN 6 CPU on older boards probably will miss some part of new features that is definitely arriving like next gen MB functionality.
INITMalcanis@reddit
I recall AMD saying something along the lines of Zen 5 introducing new CPU features and Zen 6 being more about efficiently maximising the gains from them.
However, keep in mind that in terms of revenue, consumer CPUs are merely the icing on the cake. The money is all about those Epycs. So Zen 6's "improvements" will focused on what makes it better as a server core - if that also makes them better as consumer cores, then so much the better. If not... not.
They very likely will be better as consumer CPUs, though. The Zen 5 vs Zen 5 3D difference makes it clear how much the old IMC is holding it back.
RetdThx2AMD@reddit
Aside from the other things mentioned, a new infinity fabric interconnect packaging technique like was used in strix halo. Enables higher speeds at lower power.
Professional-Tear996@reddit
Strix Halo also has almost 50% worse latency to main memory, and InFO packaging is more expensive and its manufacturing is low throughput so the final product is much higher cost.
Strix Halo using InFO is a one-off thing (likely used to fit a big GPU since SerDes takes up die space as node shrinks don't give better scaling on it) until cost and throughput improves in a way that makes it palatable for DIY.
RetdThx2AMD@reddit
InFO would not increase latency, it should reduce it because it skips the serdes step. The GPU was given priority for memory, I'd expect that would be the reason.
Professional-Tear996@reddit
AMD's implementation of InFO on Strix Halo does increase CPU <-> system memory latency.
It has been measured, for eg. in AIDA it is around 130-150 ns depending on whether the GPU is being loaded.
RetdThx2AMD@reddit
There is basically no way info is responsible for the measured increased latency. There is no where in an electrical connection for it to be introduced.
Professional-Tear996@reddit
Lol, you're saying InFO is wireless transmission of data? There absolutely are electrical connections. InFO was originally conceptualized for accelerator+HBM packaging. It is absolutely possible that latency wasn't a priority for its use, but bandwidth and power.
And CPUs typically read/write into memory at a fixed size per cycle. There has to be a way of mapping billions of read/write requests to the correct address in the memory. So there will always be some latency involved.
RetdThx2AMD@reddit
A wire does not introduce latency (other than by length, which in this case can basically be ignored*). InFO connects electrical traces from one die to another using a passive electrical connection. It introduces no more latency than a similar length connection on die. Latency is introduced using registers such as are used for a SerDes connection (at least two), where data is stored for a clock cycle. I would expect the InFO interconnect to have around 3 clocks less latency than a SerDes connection, but at the clock rates involved it is still not very significant in the grand scheme.
* a one-inch-long trace takes a small fraction of a nanosecond for electrons to traverse (less than 1/10th).
As you have picked up on, the significant latency is generally introduced on the memory and memory controller side of the design.
Professional-Tear996@reddit
Flow of electrons in a wire is orders of magnitude slower than the flow of current. You learn these things in the final two years of high school. Clearly you're talking about things you don't understand.
Even the most basic operations use two loads and one store. With 32 bytes/cycle read/write BW, that is up to 96 bytes that can be requested from or written to memory to do a single operation.
Now multiply this by the total number of instructions retired during program execution and you have potentially billions of instructions reading to/writing from memory.
And you have to keep track of all of them, which does cost latency no matter how you send those bits over from one location to another.
RetdThx2AMD@reddit
It would appear you've learned nothing from this misadventure.
Professional-Tear996@reddit
I learn by looking at the actual details on how things work, not reddit charlatans.
RetdThx2AMD@reddit
Like high latency InFO interconnects.
riklaunim@reddit
AMD is talking a lot about Fanout/"Sea of wires" and from the memory layout changes - they are probably moving to such solution for desktop as well. It won't be 1:1 Strix Halo, but similar solution. It's based on TSMC advanced packaging with lots of wires included.
Plank_With_A_Nail_In@reddit
Latency to first bit is a useless measurement, Strix Halo needs to load huge textures and 3D meshes as fast as possible. Speed to the whole data requested is the actual important measurement.
All the specs need to be taken into account focusing on just one in isolation is daft.
Professional-Tear996@reddit
I'm not talking about 3D workloads where the GPU is being used. The GPU has it easy because it is on the same die as the memory controllers. It even has MALL cache to help with its data requirements.
What is also interesting is that NVLink C2C is based on SerDes, and Nvidia doesn't have to be bound to a particular foundry for NVLink as SerDes IP is offered by everyone while only TSMC offers InFO.
And EDA tools designers like Siemens have also talked about InFO requiring chip-level tolerances instead of PCB-level tolerances when designing around it. So production throughput is always going to be a problem.
SirActionhaHAA@reddit
The latency change ain't caused by the interconnect or packaging
You'll see how wrong you are in a couple months.
Professional-Tear996@reddit
Too bad C&C didn't actually test those claims about lower latency. Stateless instant on-off doesn't mean that there is no latency involved lol.
In serdes you can encode the signal so that when you convert from serial-to-parallel or vice versa you can figure out the address mapping easily.
In fact it is this conversion/encoding that consumes the most power and costs latency in serdes. Not the actual transmission of the signal.
With InFO you still have to figure out what goes where - and that will always cost some latency.
Yeah and it runs slower LPDDR5x, not 8000 MT/s like Strix Halo does.
SirActionhaHAA@reddit
Mt/s comes at the cost of timings and latency at the same dram cost. Tighter timings and speed are harder to achieve on larger capacities which halo supports (128gb). No manufacturer is gonna ship strix halo with ultra expensive and tuned lpddr5x aimed at low latency, it doesn't have that use case
https://imgur.com/a/GMfWNeE
These are the average lpddr5x mem latencies. Just admit you're wrong and stop embarrassing yourself.
Professional-Tear996@reddit
The graph you linked doesn't tell much. It only tells that AMD has worse latencies than Intel even after accounting for 8533 MT/s on Intel vs 8000 MT/s on AMD. This is a long-standing feature of AMD because back in the days of Renoir, Robert Hallock answered a question on reddit on why Renoir is designed to loosen memory timings when the iGPU is loaded alongside the CPU.
There is no "fine-tuning" involved either other than variances due to binning of the memory IC and the memory controller on the CPU.
And the graph doesn't even include Strix Halo.
SirActionhaHAA@reddit
It showed that your bringing up of 130+ns is meaningless because that's how an average mobile sku runs. It's worse for your argument when ya consider that halo is a chiplet design and monolithic dies are similarly hitting 120-130ns
More proof that the latency ain't a result of info. Now you're just wrecking your own arguments (that info was responsible for the 130ns latency), bravo.
Professional-Tear996@reddit
Yet Intel can do sub-100 ns.
Except that it's typical for AMD mobile designs as Meteor Lake which is considered to have introduced excessively more latency being Intel's first chiplet design for client, is between Zen 5 and Zen 5c latency at the cache level and only slightly higher once you go to memory, as tested by Chips&Cheese
And Meteor Lake typically shipped with 7200 MT/s, so 10 % slower than Strix Point.
Actually, it points to the fact that AMD has to do additional optimizations that Intel doesn't which ends up compromising some other aspects of performance.
Statements which haven't been tested properly.
Even if you set aside the latency aspect, it is undeniable that InFO costs more and takes more time to implement per individual CPU. So it's cost-effectiveness will always be under question till it reaches partiy with the current flip-chip packaging.
riklaunim@reddit
It sounds like it will have a lot of structural changes. They are probably moving to Fanout instead of SERDES for communication and they have big changes for memory as as of now 2-DIMM motherboards are incompatible with Zen 6 if they use the old layout (mass refresh started few months ago for mini ITX and other 2 DIMM boards) from DIMM A1/B1 into A2/B2. This then can offer better latency and other benefits, especially when they are increasing core count per CCD (and the more cores the higher throughput needed to feed all cores with the workload). Zen 7 will probably transition this to AM6.
AnEagleisnotme@reddit
I'm going to guess Intel style stagnation
nisaaru@reddit
I "hope" for significant power consumption progress with zen 6c, 3nm io chiplet and whatever design improvement they have made.
Hayden247@reddit
Should be pretty damn big and make up for Zen 5% since it's actually skipping a node straight to N2 which is significant and resulting in some crazy clock speed rumours. Then yeah CCDs are supposed to be 12 cores now, so a 50% bump, I'd love a 10 core sku, or even 8 core Ryzen 5s if the price stays reasonable.
Qaxar@reddit
As expected. DDR6, which is AM6 big feature is still not ready for mass adoption.
Deep-Technician-8568@reddit
Makes sense. Many high speed ddr5 kits are still quite unstable. Especially if you want to run 4 sticks at high speeds. I cannot imagine how bad ddr6 will be if it's released now.
Exist50@reddit
That's not a thing DDR6 will address. The opposite, if anything.
Jiopaba@reddit
I almost wonder why they bother with four RAM slots on modern mobos. If you want more capacity you need two bigger sticks. They cheap out on everything else, who's the user here? People who don't know any better and buy 4x8?
silchasr@reddit
Maybe the cost of having 2vs4 slots isn't worth segmenting boards, plus PC's are designed around having lots of agency around picking and choosing what you want. Wouldn't it cost more for manufacturers to create 2 versions of the same product essentially just one has one less option?
Jeep-Eep@reddit
Big elephant in the room is CAMM2 - there's a good chance this DIMM debate will be fairly moot soon.
silchasr@reddit
I dunno....I think we're still quite aways from CAMM2. It'll have to be much closer to DIMM price parity. Memory is one of the last things worth upgrading since it impacts performance the least unless it's the capacity you're lacking.
Jeep-Eep@reddit
Well, one advantage of the tech is being cheaper as it only needs one PCB and interface for the same capacity.
capybooya@reddit
I think the people requiring maxing out 4 slots with the absolute largest capacity sticks will probably be fine with a smaller number of high end or production related motherboards to choose from. But the manufacturers have for many years now speculated in combining features in motherboards that very few people are interested in, so that you will have to spend more for the very few specific features that you really want, and you can't get those without also buying lots of other expensive features that they can charge you for at the same time.
Vb_33@reddit
I don't see how the consumer wins with that. The current status quo is fine.
Exist50@reddit
You still need 4 sticks for max capacity, but in general, I'm of the same mind. I really think client should just move entirely to LPCAMM2, or at least CAMM2 as standard.
Jiopaba@reddit
Sorry, let me clarify. I don't mean to say that someone who really needs serious RAM capacity can't use four sticks, just that the performance tradeoff with AMD is extreme enough that I wouldn't recommend it to anyone who doesn't really know what they're doing.
It's basically a trap, because conventional logic says if you have 16GB of RAM and you want to have 32GB you should just buy 2x8 and double it. Except with AMD that's hamstringing your performance for no reason. The folks who actually need 64/128GB of RAM and know it can certainly shop around to pick up boards that specifically support it, I just don't understand why typical consumer boards marketed for gaming workloads come with four RAM slots at this point.
Gwennifer@reddit
Then they can use the new 24gb density IC's and pick up 96gb/192gb. In fact, you can run 96GB and not even drop clocks. Even the "but I need 100GB" crowd should seriously consider a 1DPC board and 48gb DIMM's.
Also, how many applications want 192GB of memory and is only bottlenecked by single thread performance? That feels like you would rather be on Threadripper or Threadripper Pro which has a better memory controller and can do--what is it, like 1.5TB with a 2DPC board?
riisen@reddit
I have 2 asus nuc 14 pro, both with 64gb ddr5. Im about to build a new stationary now where i do need at least 96gb but will probably get 128gb.
The best would be to get a epyc platform cause i will need alot of pcie lanes aswell... but i cant afford that shit, dont just say we should get a threadripper or epyc damn have you seen those prices? I appreciate roof over my head and some food in life also.
Gwennifer@reddit
Reading can uplift many aspects of one's life.
riisen@reddit
So can anal sex.
U3011@reddit
64GB sticks came out a short while back. 128GB/256GB is now possible.
Plank_With_A_Nail_In@reddit
Some people just need lots of RAM and don't really care too much about performance.
Condemning the rest of the PC world just to appease a handful of gamers is stupid. Most PC sold never play a video game ever.
Zalack@reddit
You only need 128GB+ of RAM in desktop specialty cases like VFX, Editing 4K+ RAW video, or Scientific simulation. That’s a very small section of the market.
I edit video and program on my Desktop and 2x48 GB has been more than enough for me.
SailorMint@reddit
Wait until you hear about all those prebuilt PCs that come with only 1 stick of RAM!
Exist50@reddit
The market for >128GB of RAM in a mainstream desktop (or any desktop, for that matter), is far smaller than the gaming market. And faster RAM gives better performance in many things.
Verulamium_shore@reddit
Thing is thats the TRX50 area which from what I recall are all 4 slot despite threadripper being quad channel. WRX90s boards do run 8 slots but threadripper pro is octachannel
nanonan@reddit
TRX50 is in both 4 and 8 slot configs, see the Gigabyte TRX50 AI TOP for example.
capybooya@reddit
100%. I don't want to defend the manufacturers' extreme cost cutting and current sorry state of motherboards but 4 slots is increasingly a waste IMO.
Exist50@reddit
Agree with the sentiment. 4 slots don't make sense for the vast majority of people. Probably comes down to low incremental cost and marketing. "This board has twice the RAM slots / can fit twice the memory!" kind of thing. A sale is a sale, even if for all the wrong reasons.
Plank_With_A_Nail_In@reddit
Your experience is gaming only most PC's are not sold to play games.
Exist50@reddit
That's literally my point? Most people won't take advantage of the theoretical benefits offered by 4 DIMM slots.
kaszak696@reddit
LPCAMM2 makes no sense on desktop PCs, we don't want LPDDR memory where battery life isn't a factor. CAMM2 uses normal DDR and has more space for memory modules, it's clearly better pick for desktop motherboards.
Exist50@reddit
LPDDR is significantly faster than standard DDR, and LPCAMM has solved the traditional drawback of requiring the memory to be soldered. So now the only compelling reason for normal DDR is max capacity, the benefit of which 95%+ of people will never use.
Bloodcore911@reddit
Faster how? Do they sell 7200-8133mhz SODIMMS? Don't think so.
Pretty sure the only time LPDDR5 frequency is higher, is when its soldered, which reduces the trace length and improving the signal quality/integrity.
The same will also apply to DDR6, if soldered or using CAMM2.
Cynical_Cyanide@reddit
Err, the user is someone who wants to upgrade their RAM down the line?
I suppose it's most relevant to the beginning of a RAM generation when RAM is very expensive compared to at the end of the generation, but even still. A lot of people can barely afford a new high perf PC, and if it means they can just avoid multitasking until they can afford the other half of the RAM DIMMs, then that's probably very attractive.
I've bought 4 DIMM boards all my life, and I've always waited until some cheap used DIMMs of the same model I originally bought show up, and voila - Now I have an abundance of RAM.
Tuna-Fish2@reddit
2DPC is bad for signal integrity, and on most platforms for DDR5 it means accepting lower transfer speeds. For DDR5 and above, to upgrade ram you remove the old sticks and replace them with higher-capacity ones.
CAMM2 just formalizes this best practice, by removing 4 connectors (which you should always populate in pairs, and should always leave one pair empty) and replacing them with one.
Cynical_Cyanide@reddit
Yes I know it's hardly ideal from a purist technical standpoint. But by that logic we should forsake DIMMs and have RAM built into the mobo.
Tuna-Fish2@reddit
No, because then you can't upgrade the ram.
You can upgrade CAMM2, you just need to sell your old ram and replace it with a new module.
Cynical_Cyanide@reddit
Who's going to buy your old low capacity RAM DIMMs?
That's why four slots are good, because you can use cheap smaller capacity DIMMs additively.
Either you care about upgrading and 4 slots makes sense, or you don't and you may as well have soldered RAM for maximum perf.
Tuna-Fish2@reddit
... even if the old CAMM2 just goes to trash, you are still capable of upgrading it, it's just slightly more expensive.
Cynical_Cyanide@reddit
How is it only 'slightly' more expensive? Even if the price of RAM nosedives and you end up getting double the capacity at the same cost as your original DIMMs, then that's 2x the cost of your original set. Alternatively, you could spend say ~1.5x by simply getting a second set of smaller DIMMs, or recoup ~0.5x from resale.
I understand that you're more restricted speed wise with 4 traditional slots, but for the majority of people who aren't buying top bin RAM, does it matter? If you're buying cheap or 'sweet spot' RAM, especially early in the cycle, then you'll not have issues running a second set, which will cost you peanuts compared to buying a high speed set originally and replacing it with a modern high speed set of 2x capacity.
Blueberryburntpie@reddit
The problem is that RAM speeds absolutely tank when going beyond two sticks, and it will only get worse with the even faster DDR6.
hollow_bridge@reddit
they don't always, but a lot of people like the idea of being able to upgrade by adding more in the future, otherwise you would have to buy more expensive sticks and do something with your old ones.
Gwennifer@reddit
This is something of a fantasy because buying another 2x32 kit will drop clocks and thus execution speed on AM4/5, prompting an upgrade rather than prolonging it.
Outside of that, IC's and kits made using the exact IC don't exactly stick around forever? Like, if you try to upgrade 3 years down the line, you may not be able to buy the same capacity with the same timings anymore, or if you do they'll be as new old stock.
soggybiscuit93@reddit
Do we have actual data on the performance differences? Or what % of DIY is running RAM above JEDEC?
hollow_bridge@reddit
I do agree there is some performance loss, but if you need more capacity and you're on a budget then the performance loss is going to be less than the performance gain of the extra ram capacity.
Unless you are overclocking the timings past xmp that hasn't for a long time, it hasn't really mattered since ddr4. And overclocking the timings I've learned is actually something i would not recommend for most people as unstable ram can actually be very difficult to detect while it can destroy your data.
IMO I do agree with you that it's kind of a trap, and personally I do agree with all of your sentiment and think you should purchase with the ram you will be using when you make your initial purchase.
But humans aren't great with forward planning and budgets always exist. For example, I just recently upgraded from 2x16 3600mhz (i had purchased primarily for gaming, and am content in that usecase) to 4x16 3600mhz because I started doing stuff with AI and I realized I would have some benefits due to capacity and that it would reduce wear on my ssd. 2x32 3600 mhz would have been better. but the used extra 2x16 was $45, vs 2x32 would have been about $140. and the performance drop isn't significant enough for me to notice vs the performance gain.
kat0r_oni@reddit
"Like", sure. But honestly the last time I added more RAM to a desktop was in the 90s. And I guess thats true for the vast majority or users.
hollow_bridge@reddit
really the 90s? you had free slots? i thought computers were normally sold with their ram or IC slots maxed out back then. I've done it with every system I've had except my 2x4gb ddr3 system. I did it as recently as last week, 2x16gb>4x16gb ddr4 because I needed more for some AI stuff im doing.
Vb_33@reddit
2 sticks isn't going to get you 256GB of ram. If you need a lot of capacity for a given workload chances are you're ok losing speed to get there. Capacity or speed, pick one.
Jiopaba@reddit
Yeah, and my point is that someone who needs 256GB of RAM has a specialty use case that would justify a specialty motherboard.
The average consumer who wants an upgrade would be quite disappointed to go from 16 to 32 but lose significant performance.
RobotWantsKitty@reddit
Two extra slots are for when the first two are blocked by the CPU cooler, duh
Morningst4r@reddit
AsRock can't make enough of their 2 dimm B650m boards. Surely that's a message to manufacturers to make more 2 dimm boards for budget conscious enthusiasts
fiah84@reddit
I'm happy as a clam with 2x32GB on my 2 dimm Asrock B650 board
that is, happy with my 7800X3D, I'd have upgraded to a 9800X3D already if these boards weren't frying them by the dozen
riisen@reddit
This is why cudimm became a thing. When you have very high clock speeds and sending the clock signal from the cpu to several memory sticks it became a problem so they moved the clock to the memory module which makes it more stable and should be able to go even faster. With 5ghz the clock goes from [0 or 1] to [1 or 0] every 0.0000000002 second (200 pico seconds). So with less distance to travel the less it can get distorted or make timing errors between sticks, the timing window is very narrow at these speeds.
UsernameAvaylable@reddit
Cause they are all overclocking beyond any jedec spec...
loozerr@reddit
JEDEC has specs all the way up to DDR5-8800
GodOfPlutonium@reddit
go look at the cl latency for JEDEC and then the cl latency for the kits
loozerr@reddit
Yeah they target 14ns
mujhe-sona-hai@reddit
isn't DDR6 suppose to be CAMM2 exactly to deal with this problem?
Jeep-Eep@reddit
And I think we'll have CAMM2 AM5 within the lifespan of x870/b850, albeit maybe the wierdo Maxsun or whatever SKUs at worst
DutchieTalking@reddit
What's considered high speeds? I've got 5600mhz sticks. Would it be unstable if I added 2 more?
Deep-Technician-8568@reddit
For me high speeds means 6000mhz. If you have a cheap motherboard, adding 2 more sticks you'll most likely have issues running it at 5600mhz. You'll most likely need to drop it to 4800hmz.
DutchieTalking@reddit
Okay, thanks. Will take it into consideration if I decide to upgrade my ram total.
kazuviking@reddit
The memory is stable. The only thing making it unstable is the garbage io die and motherboard layout.
stingraycharles@reddit
Yeah even the officially vendor approved combinations are unstable. If you want to push a lot of RAM at high speeds, you’re going to have a bad time no matter what.
Jeep-Eep@reddit
There's a reason I think there's at least another gen of AM5 after this, if not two. If cached AM5 gives decent results, let Intel be the lab rat and just move up the sced for making 3d a standard feature.
Vb_33@reddit
The question is whether AM6 launches in 2028 with Zen 7 or does AMD wait till 2030 with Zen 8.
capybooya@reddit
What I'm most interested in is the DDR5 speeds of Zen6, is it too much to hope 8000+ can be stable and common/cheap enough for the release?
UsernameAvaylable@reddit
looking how people whined and bitched about DDR5 for AM5 (despite the fact that at that point DDR5 was already well established and hte prices rapidly dropping towards ddr4 levels) its really the only choice.
Acrobatic_Fee_6974@reddit
The question is whether Zen 7 will be on AM5. Given AMD have upgraded their IO die every two generations, and Zen 6 is rumored to continue this trend with a major revision, I think it's possible that Zen 7 will use the same IO die, which would imply it will not be compatible with DDR6 or PCIe 6.0, so maybe Zen 7 will be on AM5 too?
greggm2000@reddit
It could be designed to support both. Intel Alder Lake (12th gen) supported both DDR4 and DDR5, after all.
That’s the current rumor.
Weddedtoreddit2@reddit
On a 7800x3D right now. I look forward to upgrading to 11800x3D or 13800x3D in the future(or whatever the names end up being), the way people stretch out AM4 with 57/5800x3D
AMD's killing it.
Prince_Uncharming@reddit
You’re definitely set for a while. I’m on a 5700x3d and just don’t see a reason to upgrade, especially considering I got it for only $120ish and upgrading would require new ram and mobo too.
The x3d chips will have a ridiculously long life
gusthenewkid@reddit
I think this overblown, the 12900k is aging better than the 5800X3D. It doesn’t struggle with raytracing like the 5800X3D does.
Vb_33@reddit
Even if that were the case the 5800X3D runs on motherboards from the Zen 1 era in 2017 when Intel was releasing Skylake version 1.1.2. 12900k is a formidable CPU but it required a whole new platform and fast DDR5 to truly fly.
Acrobatic_Fee_6974@reddit
This is the first I'm hearing about the 5800x3D "struggling with RT", do you have a video or article to share?
Tasty_Toast_Son@reddit
I'm also quite interested to hear.
My 5800X3D does get bottomed out to 100% more often than I care to see, but I get playable framerates for now. I would still be better served by upgrading my 3080 first, methinks.
plantsandramen@reddit
5800x3d here and I'm probably waiting for the next platform to jump to
RobertISaar@reddit
Granted, much further back in time but AMD has done it before as well, the Phenom II supported both DDR2 and 3, depending on the socket/MB it was plugged into.
Exist50@reddit
Used to be some motherboards that could support either!
RobertISaar@reddit
I do seem to recall seeing some in 2009ish. 2 sockets for DDR2 and 2 for DDR3.
5heuredumat@reddit
Failed opportunity to call that shit Alderson Lake and have your BIOS say "hello, friend" each time you tap F2 during boot /s
Plank_With_A_Nail_In@reddit
Beware humans look for patterns that aren't always there. AMD's IO design process is unlikely to follow any real pattern like this.
Acrobatic_Fee_6974@reddit
Oh for sure, I could be totally off. Another user pointed out that Zen 5 was unnecessarily handicapped by using the Zen 4 IO die, so maybe that would have caused them to adopt more frequent iterations going forward. I still hope the rumors about Zen 7 being on AM5 turn out to be true.
Exist50@reddit
I think that would be reasonable. DDR6 is looking more like a 2028-ish technology, maybe even later in practice, so room enough to squeeze in Zen 7 without falling behind.
Acrobatic_Fee_6974@reddit
I read that DDR6 is going to be very expensive and probably move away from DIMMs entirely due to signal integrity constraints. So probably CAMM2 modules or something similar. Maybe consumer adoption will take a bit longer than previously for these reasons. The CEO of Silicon Motion also claimed that neither AMD nor Intel want to talk about PCIe 6 for consumer stuff yet because it's so expensive, so I think there's a chance they won't be pushing AM6 out the door too quickly.
Exist50@reddit
Where did you hear that? I certainly haven't heard anything about cost. And while I hope it does move away from traditional DIMMs, I'm not sure how that's trending in practice. The big question is what role does it and LPDDR play in a post-LLM world.
That much, I have heard. The retimer overhead alone is becoming prohibitive. Suspect we will get PCIe 6.0 in client eventually, but there's no pressing need, and unless some breakthrough is made (switch to optical?), it seems likely client will stagnate vs server.
BlueSiriusStar@reddit
One of the issues of PCIe is PAM4 modulation. One of the guys I used to work with complained that crosstalk was getting so bad with PAM4 that he had to inject noise just to denoise it later.
Acrobatic_Fee_6974@reddit
This TPU article mentions signal integrity problems that CAMM2 would solve. I honestly can't find the article I read on costs being higher than DDR5 though, it might have just been about CAMM2 modules being expensive more so than DDR6 specifically.
Exist50@reddit
In general, TPU's "news" section is garbage. This article is particularly bad.
The increase in bus speed is orthogonal to the increase in bus width.
Notebooks are pretty much all LPDDR, and LPDDR6 will be ready significantly ahead of DDR6.
This sounds made up, imo. Not aware of any server platforms planning to use DDR6 in 2027. First deployments are likely 2028 or even 2029.
Kougar@reddit
It's entirely down to whether DDR6 is ready and in the market or not. AMD has other reasons it needs to improve the IO die, reusing the IO die on Zen 5 held back the generation. I'm going to surmise reusing the Zen 6 IO die on Zen 7 will similarly detract from potential performance gains. It isn't like redesigning the IO die is some hugely prohibitive cost factor anyway, AMD does it all the time with its mobile and server SKUs.
i99990xe@reddit
There are rumors that AMD extends AM5 lifespan with Zen 7 Ryzen support: https://overclock3d.net/news/cpu_mainboard/amd-extends-am5-lifespan-with-zen-7-ryzen-support-rumour/
kwirky88@reddit
Anybody excited to see the new ways motherboard manufacturers are going to brick the next generation of cpus?
RoundGrapplings@reddit
Good move, but I wonder if older boards will need a BIOS update or end up missing some features. Hopefully performance and compatibility stay stable across all AM5 models.
INITMalcanis@reddit
Selfishly, I was low-key hoping that Zen6 would be on AM6 so I could uprgade onto a start-of-life socket, but I guess DDR6 is a pre-requisite for a new socket and it's just not there yet.
Oh well, this ol' 5800X can just keep on trucking a while yet.
Nicholas-Steel@reddit
It's not, if you look at Intel they kept changing their socket every 2 or 3 years without any notable change external to the CPU being related to it.
INITMalcanis@reddit
I didn't mean in general, I meant in the specific case of AMD superseding AM5. I should have worded it a little better.
Nicholas-Steel@reddit
The problem with buying a new CPU as soon as a new Socket is required is... early teething issues. AMD's still dealing with their 9800X3D's burning up, Intel's got their 14000 series CPU's killing themselves and both Intel & AMD are suffering from very bad Integrated Memory Controller's crippling reliability with high performance RAM.
It looks like the upcoming ZEN 6 and competing Intel CPU's are when we'll start seeing both companies tackle some of their big design issues making them the ideal generation to upgrade to if you've been holding out on like, AM4 or Intel equivalent hardware (assuming everything goes according to plan and no new major issues arise with the design changes).
INITMalcanis@reddit
9800X3D launched quite a while after AM5 debuted - I'm not really sure you can put that one down to "early teething issues". Meanwhile, to the best of my knowledge, Zen4 CPUs, including the 7800X3D, are trucking along just fine.
msolace@reddit
this is what happens with no competition, minor refreshes no real change. at least amd throwing us a bone on using same motherboards, intel would have changed sockets again. (mobo companies like selling new boards)
10/10 not hitting 7ghz, doubting anything over 6.2 /remindme
Exist50@reddit
Where did you get that from this news?
msolace@reddit
I dont call potential ddr6, and upgrade to a newer node a real change, The new node is because tsmc is reducing old node production not because its our benefit... They need to do more than add more cores/cache.
Nothing ive seen out of any producer is screaming this is the next thing. its just incremental small improvements for some money and in two years we will put the next small upgrade out.
Exist50@reddit
Lmao, absolutely not. N4 will be available for a long, long time. And AMD's leaping right to the state of the art N2.
What specifically do you want them to do, and claim this is news of them not doing?
msolace@reddit
calling it now 4% jump real world 10% useless benchmark, +300 usd, maybe some power savings.
ill take better amd idle powering, intel still beats amd on idle, amd wins everywhere else.
Not that i care about power, my powerbill is lower than my wifes tv and starbucks bill by a factor of 100...
Nicholas-Steel@reddit
They're overhauling the interconnect between cores and IOD, drastically improving power delivery and latency.
wintrmt3@reddit
The rumors are 12 core ccds, much faster interconnect, faster clocks and much lower idle power, hardly no real change.
Plank_With_A_Nail_In@reddit
Plenty of competition from Apple and ARM.
AirRookie@reddit
I’m curious zen6 if works on the 600 series chipset, but it should with a bios update right?
Azaiiii@reddit
so would that mean that a X670E mainboard would also work with these new CPUs? or only 800 series mainboards?
Exist50@reddit
Early 2027 seems quite late. Mid/late '26 seems far more reasonable. Especially since '26 is the stated timeline for Venice.
greggm2000@reddit
I’d guess it’ll depend on when N2 is ready, and where AMD fits (ie: has paid to be) in the allocation.
Exist50@reddit
N2 is supposed to be ready basically any week now, and AMD is rumored to be (one of) the very first N2 customers. I don't think that's the limiting factor.
greggm2000@reddit
N2X is what Zen 6 (Desktop) is rumored to be using, not base N2. N2X comes later, as does N2P. Ofc, that rumor may be wrong.
puffz0r@reddit
only MLID has reported that rumor. Kepler and others say n2p.
greggm2000@reddit
N2X or N2P, neither will be as soon as N2, is my point.
riklaunim@reddit
Bleeding edge nodes tend to have delays and high volume production doesn't start straight away. They have a lot of ES flying around including vendors but I doubt they will have volume available before second half 2026.
Exist50@reddit
TSMC is, in general, good about delivering when they say they will.
riklaunim@reddit
Apple had to eat delays and problems with N3 and subsequent variants
Exist50@reddit
Just the original N3. Even then, TSMC's first stated schedule was too late for the iPhone ramp, so Apple might not have been delayed at all.
Earthborn92@reddit
https://www.amd.com/en/newsroom/press-releases/2025-4-14-amd-achieves-first-tsmc-n2-product-silicon-milesto.html
SpookyKG@reddit
Who are they competing against? What's the reason to rush?
Exist50@reddit
Only the paranoid survive.
FrogNoPants@reddit
These 7 ghz block rumors seem hard to believe, so hopefully Zen6 has something interesting such as massive bandwidth increase. Wish the AMD/Intel would shorten the timeline like the ARM chips, this is like 2 years between releases..
Quatro_Leches@reddit
nah, they wont even touch 6ghz, its bs.
Exist50@reddit
They're already hitting 5.7 on N4. They're absolutely going over 6GHz next gen. On paper, the node alone could get them to mid-high 6GHz range.
puffz0r@reddit
I'd guess something like 6.4-6.5 for the top SKUs and ~6.0-6.2 for the x3ds
certainlystormy@reddit
my next build is probably going to be an amd cpu and intel gpu, jesus lmao
BlueSiriusStar@reddit
Why not an Intel CPU and Intel GPU? Nova Lake is also rumoured to use the same socket for mnay designs.
Exist50@reddit
I wouldn't be so sure about that. Sounds like we're hopefully getting RZL on the same socket. Anything beyond that seems questionable at best.
certainlystormy@reddit
suppose i could, tho it just feels like intel has stalled on cpus a bit. will wait til release for sure thooo
Exist50@reddit
Doesn't sound like there will be an Intel GPU for you to use.
MesuZako@reddit
Now we only need for a Zen 7 on AM5 confirmation.
I am on a R5 7600 so I plant to, if Zen 7 AM5 is real, to upgrade to the 12 core X3D variant once AM6 releases; and ride that till AM7.
dam0_0@reddit
Early 27 isn't that too late ? Shouldn't be Around the third quarter of 26 ?
Most likely 3dcache models will be reserved for 27 release.
metahipster1984@reddit
Hmm they only mention the 800 series being compatible. What about B650 etc?
Dreamerlax@reddit
It tracks no? AMD usually does 3 gens per socket.
Kosapt@reddit
On one hand I know what you mean and it sounds about right... on the other hand I dont think AMD had even one socket with exactly 3 generations of CPUs on it. And am4 is only 1 generation of socket so it is hard to predict trends based on only 1 sample.
Dreamerlax@reddit
Well. AM4 had Zen, Zen+, Zen2 and Zen3.
Technically 4 but I don't think Zen+ is that big of a change.
Darkomax@reddit
It technically supported Bulldozer too (whatever the last Bulldozer iteration was, Excavator or something)
dsoshahine@reddit
Didn't AM4 technically start out with Excavator APUs? So it is four, even if ignoring Zen+ (which really was more like a refresh/redo).
f3n2x@reddit
I wouldn't count Zen+ as a proper generation but absolutely would X3D, both as a performance leap as well as a massive change to the silicon configuration, even thouigh the cores themselves are the same (that's also true for the IOD between 2 and 3 though).
CrzyJek@reddit
I'm honestly surprised there were people who thought Zen6 wouldn't be on AM5...
Blueberryburntpie@reddit
Back when Arrow Lake launched and it looked like it was going to be a single CPU generation with its new socket, I've seen people argue that AMD could always pull and Intel and simply switch to a new socket for AM6 while still using DDR5.
Darkomax@reddit
Well AMD gave us AM4 owners a few scares so it's why you should not take things for granted (specifically, for the first chipset gen, because the socket support is not particularly interesting if older chipsets are left in the dust)
CrzyJek@reddit
Sheesh lol
Guillxtine_@reddit
Did Asrock commented anything about burning support of zen 6 x3d’s?
ButtonExposure@reddit
The words you're looking for is "enhanced CPU stability".
SL0WRID3R@reddit
So Zen6 != AM6 socket...
steinfg@reddit
Obviously
Beciakoteczek@reddit
Good to know — I might be building a new PC soon, so I’ll probably go with an AMD platform.
LuluButterFive@reddit
Why would Zen 6 not be supported in 800 series board? Theres no new chipset on the horizon afaik
The real questions is will zen 6 be supported in 600 series boards
riklaunim@reddit
Memory layout changes forced 2-DIMM boards refresh already. Then the more CPUs the bigger the BIOS becomes. 4-DIMM 600 should be "safe" but also AMD allegedly stopped all 600-series chipset production so there may be things behind the scene.
Deshke@reddit
huh, no zen6 next year then. thought there was a 2026 release
Albusclementor@reddit
Good news. I always say that AMD is the more future-proof platform — you don’t have to replace your motherboard with every new generation.
GamingMaster29@reddit
Will this work on my x370? 😂
Darksider123@reddit
People who bought into the platform are getting good value
ResponsibleJudge3172@reddit
You of course would expect at least 3 gens on AM5. It's zen 7 that has me doubting
Exist50@reddit
I feel like if we're getting an IO die redesign for Zen 6, they're likely to reuse it for Zen 7 and keep compatibility one more gen.
BlueSiriusStar@reddit
The IO die is indeed getting a redesign before I left. But Zen 7 may have a upgrade as well not sure what counts as a redesign anymore.
Savage4Pro@reddit
All I want is all cache 32 core cpus, no halfzies pls
MrGunny94@reddit
Can’t wait for the next 3D Cache hips to replace my 7800X3D
imaginary_num6er@reddit
Glad AsRock supports Zen 6 just as they support Zen 5 chips