AMD Zen 6 Powers "Medusa Point" Mobile and "Olympic Ridge" Desktop Processors

[Wirko]

That mean the margin on a low core count CPU aimed at the lower end will probably be very low or unprofitable.

That's a good point. However, if AMD behaves like AMD, they'll continue to sell the Zen 4 monolith for a few years more, so they have the lower end of the market covered.

 

[Zach_01]

They might be able to differentiate that 12 core CCD design and be able to replace those cores with c versions. A CCD with 4 standard cores and 8 c cores could go in laptops and also be the base for APUs. We wouldn't like it as much as a 12 standard cores CCD, but it makes sense considering Intel is selling 10 core CPUs with only 2 P cores. AMD needs to have financially viable alternatives against Intel's CPUs with a gazillion of little cores.


To cover more market segments. AMD was selling 4 core CPUs as triple cores or even dual cores back in the AM3 era, 6 cores as 4 cores, even dual cores as single cores. While we are 15 years latter, I guess this could be the case also today, with the only change to not be able to unlock those cores as we could do with AM3 CPUs. Free cores. What a nice era that was...

According to the rumors all desktop will be full Zen6 cores and for APUs there will be mixing with Zen6c cores. I don’t see the point of 6c cores on desktop. This not Intel fab with power consumption issues.
For mobile I can understand it to keep power low-er.

 

[R0H1T]

A CCD with 4 standard cores and 8 c cores could go in laptops and also be the base for APUs.

The (normal) laptop chips will almost certainly go up to 16(18?) cores. They're already at 12 with Strix point for now.

 

[john_]

I don’t see the point of 6c cores on desktop.

There shouldn't be any good reasons, other than excess quantities that can't be sold to OEMs.

 

[Punkenjoy]

To cover more market segments. AMD was selling 4 core CPUs as triple cores or even dual cores back in the AM3 era, 6 cores as 4 cores, even dual cores as single cores. While we are 15 years latter, I guess this could be the case also today, with the only change to not be able to unlock those cores as we could do with AM3 CPUs. Free cores. What a nice era that was...

AMD is a business that is there to make money. If Zen 6 low core counts doesn't make sense, they will just sell Zen 4/5 APU or CCD to cover the lower end. This is the strategy they almost always used for previous generation for cheaper part that were sold in volume.

 

[NDRE28]

I am using the Ryzen 9 9950X currently, and it's great.
I'll upgrade to the Zen 6 Ryzen 9 "10950X3D" CPU (or however they'll name it), when that will be out.

 

[RedBlackAka]

Oh God please no NPU

 

[john_]

AMD is a business that is there to make money. If Zen 6 low core counts doesn't make sense, they will just sell Zen 4/5 APU or CCD to cover the lower end. This is the strategy they almost always used for previous generation for cheaper part that were sold in volume.

They are business and if they have enough problematic dies with only 6, 8 or 10 cores working they will create products with only 6, 8 or 10 cores. But if Dell comes to them and ask a huge quantity of 6, 8 and 10 core Zen 6 CPUs, AMD will not say "Sorry, we can't cover your needs, because we only have part of the quantity you ask from us. Do you want some Zen 5 or Zen 4 instead?". That's how they will lose Dell. So they will repurpose even perfectly fine dies to cove Dell's needs. Low core count Zen 6 models make perfect sense when the order comes from an OEM and it is a huge order. Yes, your profit margins will go down in the quarter and share holders and analysts will not like it, but a huge order at lower profit margin is better than no order. That's one reason why AMD got consoles. They where willing to fulfill huge orders at probably ridiculously low profit margins.

 

[Denver]

If AMD releases a Ryzen 3 or something similar for under $200 in the next gen, it will likely be based on Zen5 APUs, probably using the Kraken architecture.

 

[Rungar]

what kind of gfx card will you need to make the most of this? Graphics card will need to be in its own box with its own power supply and dedicated 15a circuit. Good replacement for a baseboard heater..

 

[TristanX]

This is wrong speculation. AMD won't skip N3 and go straight to N2 for compute tile. I think that Zen 6 CCD will have classical 8 cores built on N3, and be released this year or early next. For packaging it is correct, they will use base die to move tiles closer and decrease latencies and power consumption, that high now for idle state.
As for N2 tile, I think it is for Zen 7 to be relaased somwhere in 2027, as demand for N2 wafers will be high (Apple, QLC, Intel), so price and availability will be difficult for AMD throughout 2026.

 

[Gameslove]

So, X3D is standard for the Zen 6?

 

[Kapone33]

What is funny is that AMD sells every chip it makes in all it's markets, with the exception of DIY. The 8700G is not as expensive as it is because AMD is being greedy but due to the fact that Handhelds have dropped a Nuclear Bomb in the PC form factor due to how well AMD APUs perform as single Gaming solutions. I will enjoy my Ally when I travel. For now my 8600G is pure fun as a HTPC. 5 GHz is sweet as f. Make sure you get an As Rock board though as 120hz support via Freesync makes Gaming a treat.

Now we get this. This will make all the people that believed the narrative about the 7900X3D and 20 nano seconds being a killer. It also seems that X3D has been a greater success than even AMD expected. It has killed the sales of the rest of them but the 7900 is still a great buy.

I also appreciate that my MB and RAM will be just fine for Zen 6. You can call me a fan boy all you want but in the CPU space AMD earns it's position with real innovation in the Space.

 

[mashie]

Let's hope they do a 24 core version, it would be a perfect upgrade from my current 5900X.

If nothing else, Conky will look sweet with 48 CPU utilisation graphs.

 

[Squared]

This is wrong speculation. AMD won't skip N3 and go straight to N2 for compute tile. I think that Zen 6 CCD will have classical 8 cores built on N3, and be released this year or early next. For packaging it is correct, they will use base die to move tiles closer and decrease latencies and power consumption, that high now for idle state.
As for N2 tile, I think it is for Zen 7 to be relaased somwhere in 2027, as demand for N2 wafers will be high (Apple, QLC, Intel), so price and availability will be difficult for AMD throughout 2026.

If AMD's standard mobile offering really does use chiplets, then 12 cores makes more sense. The alternatives are two chiplets or reducing the core count from what's currently offered. Qualcomm offers 12 cores too and on paper Intel has 14.

I've also heard that Apple might not pay for exclusive early access to N2. And I've heard that N2 is a bigger leap over N3 than N3 is over N5/N4, and that companies are considering skipping from N5/N4 to N2.

 

[Tek-Check]

As for N2 tile, I think it is for Zen 7 to be relaased somwhere in 2027, as demand for N2 wafers will be high (Apple, QLC, Intel), so price and availability will be difficult for AMD throughout 2026.

There is already 192-core EPYC on N3. The same will happen with Zen6 EPYC. Compact core top SKUs will be the first ones to move to N2, as those bring the highest revenues from hyperscalers. Do not forget that AMD designs chiplets in the same generation for more than one node.

 

[mkppo]

Zen 6 will have GMI links with enough bandwidth to not starve 12 cores, so it makes logical sense to have 2 GMI links for 24 full fat cores. I also think they will be on n3, they need a high frequency node and n3 is just getting there. The power optimized n3 version is currently in use for Turin. IOD? Samsung 4nm or whatever node doesn't really matter much as long as they design it right. Looking at Turin's IOD, i don't have too many concerns but we'll see. 3k IF and 4.5K U/Mclk should be sweet. That should be the minimum target anyway to feed 12 faster cores.

As for the inter CCD communication, strix point should be out soon and it can easily extrapolate what a exponentially faster inter CCD latency does to performance. For most productivity apps, don't expect too much. For games that cross codes? Expect a big jump but games shouldn't cross it anyway, especially with 12 cores. I'm wondering if they're going to use a variant of their fanout links that they favour in various capacities.

Too early to call it, but the performance jump should be akin to zen 2 > 3, with 1.5x cores.

 

[Wirko]

All this will require a serious step up in RAM bandwidth. LPDDR has some advantage, it wil probably surpass 9000 GT/s by the time Zen 6 comes out, but what about DDR? Any chances of MRDIMM tech coming to desktop processors?

 

[Tek-Check]

All this will require a serious step up in RAM bandwidth. LPDDR has some advantage, it wil probably surpass 9000 GT/s by the time Zen 6 comes out, but what about DDR? Any chances of MRDIMM tech coming to desktop processors?

Very unlikely. It's for data center.

 

[GerKNG]

I really hope that Zen 6 will not go down the hybrid path… I don‘t want mobile Cores in my desktop CPU. (ZenC)

 

[Carlyle2020hs]

I really hope that Zen 6 will not go down the hybrid path… I don‘t want mobile Cores in my desktop CPU. (ZenC)

Yeah. Same goes for Intel.

Slapping 2 laptop CPUs together to make a 16th Gen desktop CPU is a very cruel joke.

 

[Zach_01]

I really hope that Zen 6 will not go down the hybrid path… I don‘t want mobile Cores in my desktop CPU. (ZenC)

Nothing confirmed but as far as these rumors go the hybrid SKUs is mobile strictly.
Same source says that desktop CCDs will be 2nm.
If that’s the case I don’t see the point of using Zen6c on desktop where you have power headroom (220-230W) to implement even 24 full Zen6 cores.

 

[THU31]

It's about time they increased core counts. I really hope this means 8 cores for x600 chips, 12 cores for x700/x800X3D, and 16-24 cores for dual-chiplet CPUs.

They may have the IPC and efficiency advantage over Intel, but they're really behind in terms of core counts. And future Intel generations are expected to have some crazy configurations like 16P+32E.

 

[Tek-Check]

Zen 6 will have GMI links with enough bandwidth to not starve 12 cores, so it makes logical sense to have 2 GMI links for 24 full fat cores.

Each CCD will have two GMI links on die, but in desktop space one of them has been in use so far, to connect each CCD to IOD. If named 'GMI4', it should support speeds close to PCIe 6.0 because Venice EPYC has been leaked to introduce this interface.

Currently, one GMI3 link, a tad above PCIe 5.0 speed, supports theoretical maximum of 72 GB/s (16x36Gbps/8).
For Zen6, one GMI4 link, close to PCIe 6.0 speed, should support theoretical minimum of 128 GB/s (16x64Gbps/8).

They may have the IPC and efficiency advantage over Intel, but they're really behind in terms of core counts.

They are not "behind" because 16 cores provide overall higher performance in applications on 9950X than 24 cores on 285K. Also, 8 core 9800X3D is significantly faster in gaming than 24 core 285K. So, it's nonsense to say that they are "behind" in core count. It's not about core count.

Increasing core count to 24 on Zen6 will give 48 threads. They seem confident enough that this would be sufficient to counter alleged 52 single threaded cores on Nova Lake.

 

[pjl321]

Whilst I'm really pleased and excited that AMD is finally increasing the core count on desktop I can't help but wonder if 'only' going to 24 cores might not be enough and could AMD lose (heavily lose?) the productivity crown to Intel if (yes very big if) Intel can successfully execute on Nova Lake? Multiple leaks are pointing to Nova Lake being a 52 core monster made up of 2 core tiles each with 8 P-Cores based on Coyote Cove and 16 E-Cores based on Arctic Wolf giving a total of 48 cores plus 4 ultra low power cores (probably on the I/O die) which to me feel pointless on desktop CPUs but whatever.

So 'if' Intel does manage to get near the projected performance increases that leaks seem to be leaning towards for Coyote Cove & Arctic Wolf then I actually Nova Lake could take a sizable lead against even a 24c/48t Zen 6 CPU in productivity, I am pretty sure AMD will still hold the gaming crown fairly easily with their X3D chips plus I am really not a lover of the whole P-core/E-core thing. So many times Windows has moved a heavy workload onto my E-cores just because I've minimised the window so I can get on with something else only to realise it's progressing incredibly slowly so I have to leave it on full screen behind what I am now working on just to get it to work properly so I would likely still choose a 24 core Zen 6 CPU just because of that. But it would be great to see Intel catch up or take the lead in some scenarios if nothing else to bring some competition back to the market.

If we do some napkin maths on the possible performance and said:
1 x Zen 6 core = 100 'performance points'
1 x Coyote Cove core = 100 performance points
1 x Arctic Wolf core = 40 performance points
So Zen 6 with 24 cores = 24 x 100 = 2400
Nova Lake = 16 x 100 + 32 x 40 = 2800 (a 17% increase)

Obviously these numbers are just rough educated projections based on current leaks and rumours of Zen 6 is maybe getting a 10-15% IPC improvement over Zen 5 and Coyote Cove is possibly getting a 15-20% increase over Lion Cove. Zen 6 has multi threading of course which helps it a lot but Coyote Cove is mean to be getting a bigger IPC increase plus Arctic Wolf is rumoured to be possibly getting a massive 40-50% IPC increase over Skymont so possibly only giving them 40 'performance points' might be selling them short. But the point is it's still a bit of a guessing game at the moment and it's been a very long time since Intel managed to deliver well on anything so I'm not holding my breath they will with Nova Lake but I'm crossing my fingers it could finally get very interesting in the CPU market with Intel winner in productivity with a 52 core Nova Lake CPU but with AMD almost certainly still keeping the gaming crown!