New Zen microarchitecture details

[majord]

You can pick between 4 PCIe lanes or 2 PCIe and 2 SATA. You cant get 4 PCIe and 2 SATA.

So if we exclude graphics, for Summit Ridge you end up with either:
X370 config 1: 2 PCIe 3.0 lanes, 6 SATA and 8 PCIe 2.0
X370 config 2: 4 PCIe 3.0 lanes, 4 SATA and 8 PCIe 2.0

Compared:
Z170 20 PCIe 3.0 lanes and 6 SATA.
Z270 24 PCIe 3.0 lanes and 6 SATA.

X370 should simply have been PCIe 3.0 and more lanes. It should also have been a 28mn chipset instead of 55nm.

This is not correct. You can't have all 20/24 pci-e lanes on Z170/270 , AND the 6 SATA.
You've also failed to mention USB 3.0, in which you'd lose even more lanes if you want to use all 10. then more again to provide 3.1 via an external controller (or Alpine Ridge)

In fact if you configure Z170 in exactly the same configuration as x370 you end up with the same number of GPP pci-e lanes left over (8). Only difference is they're 3.0, and as such they can be used for more full BW NvME ports.

Obviously with Z270 you end up with 12 spare.

 

[DrMrLordX]

At 99$ to 349$ for quad to octa and one or two SKUs above that, they would ship tens of millions of units since everybody would upgrade.

Are you sure? One of the problems present in the desktop market today is that too many people are happy with what they already have.

no huge L3 cache

Right now, Zen has 8 MB of l3 per CCX. A hypothetical native quad would have 8 MB which is still pretty damn big all things considered.

The retail world is currently struggling with the fact that high-end buyers are gaining more buying power relative to mid-to-low-end buyers. Midrange buyers are getting squeezed into lower income brackets. It pays to play to the boutique market right now. Don't expect huge upgrade cycles based solely on product quality.

 

[imported_jjj]

Are you sure? One of the problems present in the desktop market today is that too many people are happy with what they already have.
Right now, Zen has 8 MB of l3 per CCX. A hypothetical native quad would have 8 MB which is still pretty damn big all things considered.
The retail world is currently struggling with the fact that high-end buyers are gaining more buying power relative to mid-to-low-end buyers. Midrange buyers are getting squeezed into lower income brackets. It pays to play to the boutique market right now. Don't expect huge upgrade cycles based solely on product quality.

This is the DIY market mostly (no integrated GPU) and it's not that people are happy with what they got, they just don't have what to upgrade too. So yeah i think they can expand the market from maybe 50 million units or less in 2016 to 80 million (as it was just a few years ago) if they offer the perf at sane prices. They would even boost the GPU market by a lot as most would be likely to upgrade the entire system,

2MB L3$ per core has been the norm in mainstream for a long time so yes it's not huge at all. Intel in the E series uses 2.5MB L3$ per core and that is costly enough in terms of area. Intel's 10 cores die is 2x the size of the 4 core APU even if the GPU is bigger than the 4 cores+cache. Ryzen is more like an APU that replaces the GPU with 4 extra cores than like the E series. That's why Intel can't compete right now at all, they need a many cores die for consumer not a server rebrand.

Not sure what's the last part about maybe you are thinking US only but China is the core market for Zen ,in volumes while in the US even the poorer have quite a lot of money. And boutique means they lose money already explained this. They don't even recup the costs if they sell hundreds of thousands or even a million(EDIT: they would actually make a little profit with 1 million units but many times less than with optimal pricing) . And ofc they would make quite a lot of $ if they ship 40 million units instead of behaving like Intel while they have no viable products in any other CPU/APU segments.

 

[DrMrLordX]

This is the DIY market mostly (no integrated GPU)

DiY market isn't that big. Zen is going to bear the most fruit in the server/workstation segment first, and then in high-end OEM PCs second. It appears that AMD will be going to the DiY segment in the same fashion that it did with k7 and k8: win the hearts of techs and sysadmins and then you get the attention of procurement people. Though I think procurement is more sophisticated thesedays. Regardless, AMD is turning the heads of people with money to burn. They will not do that with $49 CPUs.

2MB L3$ per core has been the norm in mainstream for a long time

Vishera has half the L3 of Summit Ridge despite having as many "cores" (though it only has half the number of modules, granted). The i5-6600k and 7600k only have 6 MB L3. You pretty much have to get one of Intel's top-of-the-line quads to get 8 MB L3, and to get anything more you need to go to HEDT.

Not sure what's the last part about maybe you are thinking US only but China is the core market for Zen

I'm sure many people will be buying Summit Ridge products in China for business, and boutique buyers will be buying it there as well, but Lisa Su has specifically said that AMD is not interested in being the "discount" brand. China is a perfect exemplar of the shift in buying power globally towards the wealthy. AMD is not banking on mass sales of $49 Zen-based chips, certainly not in H1 2017. I do not think they will even sell Zen at $99.

 

[imported_jjj]

DiY market isn't that big. Zen is going to bear the most fruit in the server/workstation segment first, and then in high-end OEM PCs second. It appears that AMD will be going to the DiY segment in the same fashion that it did with k7 and k8: win the hearts of techs and sysadmins and then you get the attention of procurement people. Though I think procurement is more sophisticated thesedays. Regardless, AMD is turning the heads of people with money to burn. They will not do that with $49 CPUs.
Vishera has half the L3 of Summit Ridge despite having as many "cores" (though it only has half the number of modules, granted). The i5-6600k and 7600k only have 6 MB L3. You pretty much have to get one of Intel's top-of-the-line quads to get 8 MB L3, and to get anything more you need to go to HEDT.
I'm sure many people will be buying Summit Ridge products in China for business, and boutique buyers will be buying it there as well, but Lisa Su has specifically said that AMD is not interested in being the "discount" brand. China is a perfect exemplar of the shift in buying power globally towards the wealthy. AMD is not banking on mass sales of $49 Zen-based chips, certainly not in H1 2017. I do not think they will even sell Zen at $99.

The DIY market is what Zen targets and is is plenty huge (considering how little AMD sells now). Folks that buy it in a pre-built PC are few. Workstation, real workstation is pretty low volumes. Server ,maybe they address part of it with the same die but that is yet to be confirmed and it would be decision forced by their limited resources and minimal existing share.Zen desktop could finance a proper Zen+ server offering.
The cache on some of Intel's SKUs is artificial,they have the cache but they disable it and lots of SKUs have the full cache enabled but in high end consumer desktop they are extra crippled to make the top SKUs more appealing.

My suggested prices are reasonable not aggressive and far from cheap and made an extra effort to make that clear in my comment by even looking at margins.Even 349$ is a very high a bit of a ripoff but people are used to that. China is huge for the DIY market even if we don't know it. If you look at Asus and Gigabyte, about half of their mobo shipments are in China - they ship 16-17 million units each worldwide at this point. You seem to have a bit of a thing for boutique but really?
This is not aimed at OEMs, businesses, workstation and even server,it's about the enthusiasts and the best definition for that is the DIY market, they are enthusiastic enough to build their own box.

In the end ,you sell it where you make the most money and the most money is by serving the entire market except sub 99$ and that's only for now. As already mentioned, even the die is designed for this.They don't waste power and area on things that add little perf and convenience,like Intel's E line. APUs would come later and offer more options in the same price bands.

Edit: in OEM margins and ASPs would be much lower than what they would get in retail, retail is a few times more appealing.

 

[beginner99]

At 99$ to 349$ for quad to octa and one or two SKUs above that, they would ship tens of millions of units since everybody would upgrade.
They could even go to 49$ for quads but it is likely that ,this year, most that buy bellow 99$ would make an effort and get the quad. Next year,when Zen+arrives,they can drop the prices for this gen to 49$.

No 4 chans memory ,no huge L3 cache, nice IO but not excessive.They are gonna hit Intel where it hurts, across the board and serve every customer that doesn't need a silly integrated GPU.
If Zen delivers and AMD has sufficient supply, it's gonna be weird. The massive upgrade cycle it will cause....

That is wishful thinking. Have you looked at the Kaby lake release? Yes we all complain about the 7700k only being a higher clocked 6700k. But the real deal with kaby lake is once again mobile and relevant for this thread in the budget sector.

DIY is a tiny niche of the total sector. Not having an iGPU is actually a pretty serious handycap for OEM sector were the real high volume gets shipped. Intel now offers a hyper-threaded pentium with 3.5 Ghz clocks at $64 and an iGPU (only GT1 but good enough for office, and media use). AMD can't compete with that with a 4-core Zen as it lacks the iGPU. They need the APU for this sector. If ST performance of Zen is good-enough and the 4-core variant can clock a bit higher like 3.7 with 4 turbo, yes it will be great for budget gaming builds but that will not move huge volumes. For huge volumes you need to be in OEM market be it desktops or server. Also you don't want to market your new, high-end server product in the same segment as Pentiums and Celerons.

Therefore I have my doubts they will actually offer a cheap 4-core variant or one at all. Probably depends on yields as well. It it performs demand will be high and you want to sell as many chips as you can as 8-core versions without manually turning off functional cores and sell them as quads. If they do offer a quad it will go against lowest i5, highest i3 pricing so around $150.

 

[imported_jjj]

That is wishful thinking. Have you looked at the Kaby lake release? Yes we all complain about the 7700k only being a higher clocked 6700k. But the real deal with kaby lake is once again mobile and relevant for this thread in the budget sector.
DIY is a tiny niche of the total sector. Not having an iGPU is actually a pretty serious handycap for OEM sector were the real high volume gets shipped. Intel now offers a hyper-threaded pentium with 3.5 Ghz clocks at $64 and an iGPU (only GT1 but good enough for office, and media use). AMD can't compete with that with a 4-core Zen as it lacks the iGPU. They need the APU for this sector. If ST performance of Zen is good-enough and the 4-core variant can clock a bit higher like 3.7 with 4 turbo, yes it will be great for budget gaming builds but that will not move huge volumes. For huge volumes you need to be in OEM market be it desktops or server. Also you don't want to market your new, high-end server product in the same segment as Pentiums and Celerons.
Therefore I have my doubts they will actually offer a cheap 4-core variant or one at all. Probably depends on yields as well. It it performs demand will be high and you want to sell as many chips as you can as 8-core versions without manually turning off functional cores and sell them as quads. If they do offer a quad it will go against lowest i5, highest i3 pricing so around $150.

I am not sure that you have considered the math at all and AMD's current revenue.
This CPU only Ryzen thingy is not for OEM, i have never claimed it is.It is for us,for the DIY market mostly and it is a huge market if you think about it.
If you look at JPR stats for discrete GPUs in desktop, last year will end up at likely above 45million units. PCs with CF and SLI are a small percentage. There is a huge market in units for people that don't need integrated GPUs.
That's in units but there is a lot more than that to this. ASPs and margins are fantastic compared to OEM sales.In laptop they would get 70$ ASP and w/e volumes- Intel gets higher ASPs today because there is no competition and they milk the low power segment with a small die at high price.
In DIY it's much much easier to gain share as the consumer is better informed and more rational then the average consumer.These buyers ask for advice, read reviews.And Intel has left the door wide open, AMD can easily offer twice the cores with great margins and ASPs (up to 349$is actually great). In this market , given the context AMD can get 50% share, in laptop they need time ,marketing and they are lucky if they get to 20% share in 2018. This year and next year, CPU with,no GPU can generate much much more revenue and income that APUs.This is the lowest hanging fruit and the biggest opportunity as Intel has been severely mistreating the segment for so many years. They can also boost the market with a huge refresh cycle, boost GPU sales with it, gain billions in mind share.

If the octa cores clock to 4GHz or better ,the quads will do much better. Gaming doesn't scale with clocks ,some do but some scale with cores and the direction is towards more cores. You also don't understand what kind of hardware gamers have.very few pay 349$ for a CPU or more than 250$ for a GPU, globally. AMD won't lock the CPUs, might not even cripple them- Intel disables heaps of feature. And lets not forget that AND has the "as high as the cooling allows it" turbo. If you want to pay 300+ for Kaby with integrated GPU for gaming, instead of 179$ or199$ for a 4.5GHz (or higher with good cooling) quad Zen, it's not very wise but what do i care.
The Pentium you mention is dual core and yes Ryzen doesn't try to compete with that, that's the entire point for this platform and that gives it such a huge potential.

As for the remaining market, they will have APUs,from dual core 5W, maybe even lower than 5W since Zen seems to exceed expectations.And it is actually really encouraging that Vega is looking good.They already had a GPU advantage over Intel and Vega might make it a huge lead.Plus AMD actually works with game developers and that matters.The problem here is that the consumer market that remains (excluding the market zen addresses) is not that big- and do remember low ASPs and moderate to less than that margins. So the market in terms of $ is so much smaller and it's more difficult to reach the consumer if you don't have billions to spend on marketing every quarter. This market has been declining hard, the consumer PC. The business segment holds much better and that's the other segment APUs will address. Very high volumes but not great margins and ASPs and not that easy to gain share since Intel is so well established and AMD has been rather week in this segment.

You also seem to think that they won't have a native 4 cores Ryzen die but the hints so far point to such a die being launched from day one and as i said, any price band they don't address,is business left on the table and it just feeds Intel. It would be bad business.

 

[itsmydamnation]

The DIY market is what Zen targets and is is plenty huge (considering how little AMD sells now). Folks that buy it in a pre-built PC are few. Workstation, real workstation is pretty low volumes. Server ,maybe they address part of it with the same die but that is yet to be confirmed and it would be decision forced by their limited resources and minimal existing share.Zen desktop could finance a proper Zen+ server offering.

What are you talking about, Zen is a Server designed SOC. It is design with a distributed memory controller/ north bridge architecture. Zen server is designed for 32core/64 thread/8 memory channel/~64 pci-e lanes a socket. This is done via 4 way MCM packaging using a coherent memory interconnect. This spec's either meet or exceed skylake-EP, what is also interesting is that Xeon V5 uses 3647 and Naples is 4k. You then have the server APU which is 2x Zeppelin and 1x Vega on a MCM package, both Vega and Zeppelin were designed from scratch with this in mind.

Also CPU's and SOC architectures are multi year development efforts, the revenue from Zeppelin isn't changing any short to mid term plans ( it can very much help on development of the existing planned products) but if AMD are doing yearly release as said by papermaster and the Zen design has a 4 year life cycle sa said by papermaster then all of Zen's follow on's are already at various stages of design.

 

[imported_jjj]

What are you talking about, Zen is a Server designed SOC. It is design with a distributed memory controller/ north bridge architecture. Zen server is designed for 32core/64 thread/8 memory channel/~64 pci-e lanes a socket. This is done via 4 way MCM packaging using a coherent memory interconnect. This spec's either meet or exceed skylake-EP, what is also interesting is that Xeon V5 uses 3647 and Naples is 4k. You then have the server APU which is 2x Zeppelin and 1x Vega on a MCM package, both Vega and Zeppelin were designed from scratch with this in mind.
Also CPU's and SOC architectures are multi year development efforts, the revenue from Zeppelin isn't changing any short to mid term plans ( it can very much help on development of the existing planned products) but if AMD are doing yearly release as said by papermaster and the Zen design has a 4 year life cycle sa said by papermaster then all of Zen's follow on's are already at various stages of design.

Actually that is mostly speculation (how will they address the server) and i was clearly talking about the 8 core die that addresses desktop.No fat at all. The server leaks and rumors have been a bit contradictory and i need more to be certain. If they do have 4x to reach 32 cores, that's obviously less than ideal even with advanced packaging and the Infinity fabric and points to pretty much what i said.that they didn't had the resources to do their best in addressing server. Not that i am saying that Zen will be a dud in server, don't think we have the info to even speculate on that. Just like they still have work to reach the full potential in addressing all the segments in GPU (gaming, HPC, machine learning).

Note that Zen enabling investments in Zen+ based products (not the core, the different SoCs), doesn't mean that the move wasn't planned years ago.It doesn't imply a change in strategy now, at this point. AMD planed for revenue growth enabled by Ryzen and for that revenue to enable investments in certain key areas.
They can't go from 0 to 100% in all segments in a single year -the 0 is hyperbolic don't get stuck on that
So far it looks like they have turned the ship around in a spectacular way but there is some work left to do.

EDIT: the sever market this year might be towards 25 million units CPUs at an ASP a bit above 600$- just an approximation. So that's a 15 billion market with great margins but how much share can they realistically get even if some very large customer jumps ship and AMD grabs share in bulk. Would be great for them if they get a lot but maybe 10% would be ok for now. Oh and i am mixing the Ryzen and Zen terms instead of using Summit Ridge when talking about this desktop platform, that's my bad, should have been more precise.

 

[itsmydamnation]

Actually that is mostly speculation (how will they address the server)

No its not, You can find details in many linked in profiles, you can find leaks where everything else so far as been 100% correct. There is then the compile patches which describe how the L3 is arranged for upto 32cores and it also aligns. I have also had discussions with a person from an OEM who is Intimately aware of Zen's server configuration and what we where talking about would not make sense without >2 mcm chips.

and i was clearly talking about the 8 core die that addresses desktop.No fat at all. The server leaks and rumors have been a bit contradictory and i need more to be certain.

It has lots of fat if it was just a desktop part, maybe you should go look at the die shot and then see the massive amount of I/O in the bottom left (just as big as the memory interfaces and they add up to 288 pins). Then there are things like it has ECC on the L1 cache, why would they waste transistors and power on ECC for consumer part where ECC has ZERO support. The Zepplin SOC also has multi 10gbe interfaces, why if its just consumer?

If they do have 4x to reach 32 cores, that's obviously less than ideal even with advanced packaging and the Infinity fabric and points to pretty much what i said.

How is it less then ideal? what makes it less then ideal? From a Moores law perspective its very attractive. From a binning perspective its very ideal. We dont know how the cache coherency works inter CCX but it already has to exist for a Zeppelin SOC because there are two CCX's.

The two area's that aren't ideal would be inter chip latency and the extra power but without knowing what AMD is using for coherency ( is it still MOESI or something new) its hard to know how much it really costs, but it has the befit vs Xeon (at least V4) that the local L3 is much faster.

that they didn't had the resources to do their best in addressing server.

AMD have always used MCM server parts and on the GPU side they plan to do something simlar with Navi so it looks like a deliberate direction. It also helps minimize the impact of process variance.

Not that i am saying that Zen will be a dud in server, don't think we have the info to even speculate on that. Just like they still have work to reach the full potential in addressing all the segments in GPU (gaming, HPC, machine learning).

From the same discussion as i said before they claimed that the 32core Zen parts performance is "very good". AMD themselves said they can complete on performance with the Xeon E5-2699 V5 on performance ( a 32core 64thread chip)

Note that Zen enabling investments in Zen+ based products (not the core, the different SoCs), doesn't mean that the move wasn't planned years ago.It doesn't imply a change in strategy now, at this point. AMD planed for revenue growth enabled by Ryzen and for that revenue to enable investments in certain key areas.

The point is the next Zen if it is coming in ~12months AMD would already/very close to have prototypes back, The Zen after that has 9 month until it needs to be sent to the fabs etc etc. We already know AMD's R&D funding because its published and its been stable so unlikely to see and more SOC's then we currently do for at least the next two successors.

They can't go from 0 to 100% in all segments in a single year -the 0 is hyperbolic don't get stuck on that
So far it looks like they have turned the ship around in a spectacular way but there is some work left to do.

AMD said they can address 80% of the server market with Zen/snowy owl/Naples. the missing 20% is because Zen's design can't go past 2 sockets. that rules out the big DB/ERP systems etc.


edit: wrote wrong coherency protocol.

 

[iBoMbY]

You might be a bit confused.
Ryzen should offer 32 PCIe lanes for dual GPU at x16, not sure if we have confirmation but that's what we assume.

That is wrong. Ryzen has 24 PCIe 3.0 lanes. 4 are used for the chipset, that leaves 20. 4 can be used for SATA or NVMe, and 16 (or 8/8) for GPUs.

 

[coercitiv]

That is wrong. Ryzen has 24 PCIe 3.0 lanes. 4 are used for the chipset, that leaves 20. 4 can be used for SATA or NVMe, and 16 (or 8/8) for GPUs.

So when talking storage performance, can Zen use up to 8 PCIe 3.0 lanes for data transfer? (4 via chipset, 4 directly)

 

[iBoMbY]

So when talking storage performance, can Zen use up to 8 PCIe 3.0 lanes for data transfer? (4 via chipset, 4 directly)

Yes, pretty much. Basically enough for two fast NVMe SSDs, or one fast NVMe SSD and four SATA drives. If you use a fast NVMe SSD and SATA, plus other PCIe devices, at the same time on the X370 that's probably going to cost performance.

Maybe you could use the second x16 slot (running at x8) for additional I/O, when limiting the GPU to x8.

 

[bjt2]

What are you talking about, Zen is a Server designed SOC. It is design with a distributed memory controller/ north bridge architecture. Zen server is designed for 32core/64 thread/8 memory channel/~64 pci-e lanes a socket. This is done via 4 way MCM packaging using a coherent memory interconnect. This spec's either meet or exceed skylake-EP, what is also interesting is that Xeon V5 uses 3647 and Naples is 4k. You then have the server APU which is 2x Zeppelin and 1x Vega on a MCM package, both Vega and Zeppelin were designed from scratch with this in mind.

Also CPU's and SOC architectures are multi year development efforts, the revenue from Zeppelin isn't changing any short to mid term plans ( it can very much help on development of the existing planned products) but if AMD are doing yearly release as said by papermaster and the Zen design has a 4 year life cycle sa said by papermaster then all of Zen's follow on's are already at various stages of design.

Top INTEL server SKUs has 8 channel memory? Or 6?

 

[bjt2]

Actually that is mostly speculation (how will they address the server) and i was clearly talking about the 8 core die that addresses desktop.No fat at all. The server leaks and rumors have been a bit contradictory and i need more to be certain. If they do have 4x to reach 32 cores, that's obviously less than ideal even with advanced packaging and the Infinity fabric and points to pretty much what i said.that they didn't had the resources to do their best in addressing server. Not that i am saying that Zen will be a dud in server, don't think we have the info to even speculate on that. Just like they still have work to reach the full potential in addressing all the segments in GPU (gaming, HPC, machine learning).

Note that Zen enabling investments in Zen+ based products (not the core, the different SoCs), doesn't mean that the move wasn't planned years ago.It doesn't imply a change in strategy now, at this point. AMD planed for revenue growth enabled by Ryzen and for that revenue to enable investments in certain key areas.
They can't go from 0 to 100% in all segments in a single year -the 0 is hyperbolic don't get stuck on that
So far it looks like they have turned the ship around in a spectacular way but there is some work left to do.

EDIT: the sever market this year might be towards 25 million units CPUs at an ASP a bit above 600$- just an approximation. So that's a 15 billion market with great margins but how much share can they realistically get even if some very large customer jumps ship and AMD grabs share in bulk. Would be great for them if they get a lot but maybe 10% would be ok for now. Oh and i am mixing the Ryzen and Zen terms instead of using Summit Ridge when talking about this desktop platform, that's my bad, should have been more precise.

Fiji connects 5 dices, with 4x1024 memory busses together with the interposer. Bus width will not be a problem. Dices number neither. The problem can be the bus clock: HBM2 have 2GT/s. Even if for the interconnect fabric they can't go higher than that (but i doubt since with GDDR5 they went off chip at 7GT/s/pin), they could go with 1024-2048 bit buses...

 

[itsmydamnation]

Top INTEL server SKUs has 8 channel memory? Or 6?

I believe E5- V5's have 6 channels. LGA2011 E5 V5's have 4 channels.

 

[coercitiv]

Yes, pretty much.

And how does that compare with what Intel's HEDT has to offer?

 

[CatMerc]

And how does that compare with what Intel's HEDT has to offer?

Something between Intel's 1151 and 2011 platforms. Personally think this is an excellent position for AMD, as the 2011 platform has a ton of things that honestly don't make much of a difference for non servers, but they increase prices for every motherboard. Meanwhile the 1151 platform has just shy of what I would like to see in terms of I/O.

Cheap motherboards with plenty of I/O for consumer applications.

 

[iBoMbY]

And how does that compare with what Intel's HEDT has to offer?

The 2011 CPUs have, as far as I know, 28 or 44 lanes plus DMI 3.0 to the chipset, which is about the same as one x4. So it's 32/48 vs. AMD's 24.

 

[coercitiv]

The 2011 CPUs have, as far as I know, 28 or 44 lanes plus DMI 3.0 to the chipset, which is about the same as one x4. So it's 32/48 vs. AMD's 24.

Well, AFAIK current 2011 CPUs and X99 chipset are still on DMI 2.0, which puts the chipset at half throughput. HEDT is bound for an update soon after Zen though.

 

[ShintaiDK]

What are you talking about, Zen is a Server designed SOC. It is design with a distributed memory controller/ north bridge architecture. Zen server is designed for 32core/64 thread/8 memory channel/~64 pci-e lanes a socket. This is done via 4 way MCM packaging using a coherent memory interconnect. This spec's either meet or exceed skylake-EP, what is also interesting is that Xeon V5 uses 3647 and Naples is 4k. You then have the server APU which is 2x Zeppelin and 1x Vega on a MCM package, both Vega and Zeppelin were designed from scratch with this in mind.

Also CPU's and SOC architectures are multi year development efforts, the revenue from Zeppelin isn't changing any short to mid term plans ( it can very much help on development of the existing planned products) but if AMD are doing yearly release as said by papermaster and the Zen design has a 4 year life cycle sa said by papermaster then all of Zen's follow on's are already at various stages of design.

Are you sure its not 32 PCIe and quadchannel per socket?

And the fabled server APU was nothing but vague slides 2 years ago.

 

[moonbogg]

32pci lanes is enough for me since I'm done with multi GPU. That's 16 for a single GPU and plenty left over for SSD's, sound cards and all that other hoopla. EK is ready to go with mounting brackets and everything. I think they are going to sell a crap ton of these chips.

 

[Dresdenboy]

Are you sure its not 32 PCIe and quadchannel per socket?

And the fabled server APU was nothing but vague slides 2 years ago.

Are you talking about SP3 or SP4 chips? The CERN leak said 8 channels. Ian Cutress found multiple hints for 8 channels per socket on the Naples board.

 

[mohit9206]

Wow i didn't know intel was introducing Hyper threaded Pentiums with kaby Lake. Wouldn't this lower the sales of i3? Why buy an i3 when $64 Pentium makes much better value for money and allows the person to buy a faster gpu. Pentium G4560 + Gtx 1050 is good cheap combination.

 

[krumme]

Wow i didn't know intel was introducing Hyper threaded Pentiums with kaby Lake. Wouldn't this lower the sales of i3? Why buy an i3 when $64 Pentium makes much better value for money and allows the person to buy a faster gpu. Pentium G4560 + Gtx 1050 is good cheap combination.

Naa its extremely limited and intel knows what they are doing. People buy a name and a brand. Its like 0.01% that knows whats actually the perf of the cpu.
Core i3 i5 i7 is name that at least a few % knows the broader meaning of.Thats why amd name their cpu sr3 sr5 sr7 and it goes to show how they will segment and price the products. They will probably try to get a premium position by adding like 10-20% to the price of the core equivalent. Pretty smart move imo. Dont know if it will succeed but they are gunning for the core line segments.
If top end zen ends faster than 8c bwe we might see some extreme Intel like pricing for the top bin and a side effect for next in line but Intel pricing is where its heading.