Discussion Speculation: Zen 4 (EPYC 4 "Genoa", Ryzen 7000, etc.)

[Vattila]

Except for the details about the improvements in the microarchitecture, we now know pretty well what to expect with Zen 3.

The leaked presentation by AMD Senior Manager Martin Hilgeman shows that EPYC 3 "Milan" will, as promised and expected, reuse the current platform (SP3), and the system architecture and packaging looks to be the same, with the same 9-die chiplet design and the same maximum core and thread-count (no SMT-4, contrary to rumour). The biggest change revealed so far is the enlargement of the compute complex from 4 cores to 8 cores, all sharing a larger L3 cache ("32+ MB", likely to double to 64 MB, I think).

Hilgeman's slides did also show that EPYC 4 "Genoa" is in the definition phase (or was at the time of the presentation in September, at least), and will come with a new platform (SP5), with new memory support (likely DDR5).

What else do you think we will see with Zen 4? PCI-Express 5 support? Increased core-count? 4-way SMT? New packaging (interposer, 2.5D, 3D)? Integrated memory on package (HBM)?

Vote in the poll and share your thoughts!

 

[Saylick]

Well, if it wasn't obvious enough why Zen 4, and even Zen 3D, isn't shipping to consumers anytime soon... AMD has been focusing on making money for shareholders over everything else. TSMC's last earnings projected 30% more revenue YOY, and assuming that means 30% more sold wafers, it's not hard to imagine where the bulk of those sold wafers to AMD are going.

https://twitter.com/x/status/1481016175140114432

 

[nicalandia]

Well, if it wasn't obvious enough why Zen 4, and even Zen 3D, isn't shipping to consumers anytime soon... AMD has been focusing on making money for shareholders over everything else. TSMC's last earnings projected 30% more revenue YOY, and assuming that means 30% more sold wafers, it's not hard to imagine where the bulk of those sold wafers to AMD are going.

https://twitter.com/x/status/1481016175140114432

Is that a bad thing? I would say take away as much % of market share of Intel's Server/HPC market as possible.

 

[jpiniero]

Is that a bad thing? I would say take away as much % of market share of Intel's Server/HPC market as possible.

Imagine if yields were good enough for Epyc between the regular and the F to the point where AMD cancels Raphael.

 

[RnR_au]

Imagine if yields were good enough for Epyc between the regular and the F to the point where AMD cancels Raphael.

I believe this was why Cutress was last seen rocking a Zen 4 EUV machine at TSMC to hurry Raphael along...

 

[tomatosummit]

Is that a bad thing? I would say take away as much % of market share of Intel's Server/HPC market as possible.

I would say increasing prices are a bad thing.

 

[Markfw]

I would say increasing prices are a bad thing.

I would say until AMD gets 50% market share in servers, anything goes to help them to that goal. Once there, I want competition ! I want to see Intel fight back ! Right now they can't in servers.

Edit: I hope the same for desktop, Ryzen 4. I want to see 50% market share.

Can you imagine the competition when AMD and Intel both get to 50% ? Can you say price war ? (assuming they are both close in performance and efficiency)

 

[tomatosummit]

I would say until AMD gets 50% market share in servers, anything goes to help them to that goal. Once there, I want competition ! I want to see Intel fight back ! Right now they can't in servers.

Edit: I hope the same for desktop, Ryzen 4. I want to see 50% market share.

Can you imagine the competition when AMD and Intel both get to 50% ? Can you say price war ? (assuming they are both close in performance and efficiency)

What, why would anyone other than a shareholder give a damn about amd's marketshare?
The price increases are amd abusing their position as they both have the premium product and a supply constraint now. Regular thinking would be reducing pricing to drive demand to gain market share, the opposite of that is happening right now.
No business today (across a lot of fields) has any interest in starting a price war when they can just say "supply problems"

 

[DisEnchantment]

I would say until AMD gets 50% market share in servers, anything goes to help them to that goal. Once there, I want competition ! I want to see Intel fight back ! Right now they can't in servers.

Oh they have competition, they managed to stave off mass hyperscaler exodus from x86 for now. Intel should be grateful to AMD on that one.

Regular thinking would be reducing pricing to drive demand to gain market share, the opposite of that is happening right now.
No business today (across a lot of fields) has any interest in starting a price war when they can just say "supply problems"

Because they need to have healthy GM to invest in R&D. Semiconductor industry is capital intensive.
Just having higher revenue, like MTK for instance, is not healthy, without good GM means it is not healthy business. It means the company doesn't have much innate value from its own IP, but has lots of external dues (like IP licensing) to be paid. There is a reason NVDA stock is much more valuable than MTK.
You cannot say "supply problems" to corporate customers, they know. They also don't buy on the spot, from the market. The book order in advance, directly from AMD without any distributor and/or intermediaries.

Besides that, AMD's suppliers raised prices, they can't be shouldering the cost on their own.
And it is for corporate customers, can't care less if they pay anyway.

 

[Mopetar]

Well, if it wasn't obvious enough why Zen 4, and even Zen 3D, isn't shipping to consumers anytime soon... AMD has been focusing on making money for shareholders over everything else. TSMC's last earnings projected 30% more revenue YOY, and assuming that means 30% more sold wafers, it's not hard to imagine where the bulk of those sold wafers to AMD are going.

https://twitter.com/x/status/1481016175140114432

You probably missed the news release where TSMC was raising prices. I would imagine if there are shortages of other components or supplies, those have gone up as well. AMD is just passing those costs along just like anyone upstream of them did with their costs. If they can raise prices with a "take it or leave it" attitude it's pretty obvious they have more people who want their chips than they can possibly supply.

I would say until AMD gets 50% market share in servers, anything goes to help them to that goal. Once there, I want competition ! I want to see Intel fight back ! Right now they can't in servers.

Edit: I hope the same for desktop, Ryzen 4. I want to see 50% market share.

Can you imagine the competition when AMD and Intel both get to 50% ? Can you say price war ? (assuming they are both close in performance and efficiency)

It would take a lot for AMD to get to 50%. I'm not even sure if they can in the near future even if Intel completely screwed up and had their own Bulldozer moment. AMD doesn't have access to enough wafers to get to 50%. They'd need to be buying wafers off of other companies in order to do so or at least any new capacity that TSMC adds.

Really, neither company needs to be any market share to improve competition. It's just a matter of having products that are competitive and enough excess production capacity to drive prices down. Right now there's just too much demand to really force prices down. Let's just be glad the CPU market doesn't look like the GPU market.

Incidentally the best way for AMD to get to 50% market share is to do what they're doing and target the highest margin spaces. That gives them more revenue and the ability to buy more wafers. Of course then people complain about AMD not caring about consumers, but there's no pleasing some people.

 

[DisEnchantment]

Really, neither company needs to be any market share to improve competition. It's just a matter of having products that are competitive and enough excess production capacity to drive prices down. Right now there's just too much demand to really force prices down. Let's just be glad the CPU market doesn't look like the GPU market.

Incidentally the best way for AMD to get to 50% market share is to do what they're doing and target the highest margin spaces. That gives them more revenue and the ability to buy more wafers. Of course then people complain about AMD not caring about consumers, but there's no pleasing some people.

Yeah, just need a market share above a certain point to make enough profit and then product positioning to get better GMs

But Intel and Samsung should have done better on the foundry front. Seeing the progress of the entire industry depending on the supply chain concentrated in a small Island bothers me.

Fingers crossed for Intel 4 and 3GAE.
Next year when I buy a laptop when travelling becomes normal again I hope to be able to chose a low wattage Intel 4 or AMD N5 laptop chip.

 

[Saylick]

You probably missed the news release where TSMC was raising prices. I would imagine if there are shortages of other components or supplies, those have gone up as well. AMD is just passing those costs along just like anyone upstream of them did with their costs. If they can raise prices with a "take it or leave it" attitude it's pretty obvious they have more people who want their chips than they can possibly supply.

Ahh, that's a good point. Not all revenue gains has to come from more sold wafers, which makes sense since TSMC probably doesn't have more capacity anyways.

 

[amrnuke]

This "supply shortage" is certainly interesting. I can walk into any Microcenter and buy a 6900XT GPU, or a 5950X CPU, or a 5700G or 5600G, any day of the week. I know it's more complex than that, but it's hilarious to see the supply chain issues as an excuse for higher prices. The CPU prices are reasonable. The GPU prices are higher because AIBs know some people will pay those prices. So they are willing to let product sit on the shelves until the next mining bubble hits. It's not (just) a supply chain issue. It's a priority issue.

 

[Tarkin77]

Igor in his latest Video is saying, AM5 and Raphael coming in Q3 ... so a Comutex reveal seems in the cards

(its german, sorry)

 

[DrMrLordX]

Is that a bad thing? I would say take away as much % of market share of Intel's Server/HPC market as possible.

Raising prices increases revenue share. Raising volume increases market share. Not the same thing. Besides, as noted above, it appears as though TSMC's customers are paying more for wafers now so . . . what does anyone honestly expect?

Fingers crossed for Intel 4 and 3GAE.

Intel 4 is low-volume and may never be used for anything outside of Meteor Lake and Granite Rapids. 3GAE has (allegedly) been pushed so far back that it will never reach the market and instead be an internal process (3GAP will eventually take its place).

 

[DrMrLordX]

(its german, sorry)

There's an English article linked in the comments section:

AMD’s Socket AM5 v2 in detail - What is better in Ryzen than in Intel’s Socket LGA-1700 for Alder Lake | Exclusive | igor´sLAB

Today I turn for the first time in detail to AMD’s upcoming Socket AM5 (version 2), as it will be used for the upcoming Raphael CPUs (Ryzen 7000). AMD consequently changes from the PGA (Pin Grid…

www.igorslab.de

It doesn't have all the info from the video though.

 

[SteinFG]

Seeing as am4 and am5 chip's substrate is the same size, I can already foresee a ton of inexperienced builders jam an am4 cpu pins into an lga grid on am5.

 

[szrpx]

Difficult to say much about those without running some simulations for these cutting edge process/SoCs. I bet nobody is going to share such things for obvious reasons.
Historically you'd be calculating power as P ∝ C*V2*f. But f also is a nonlinear function of V. Reducing V sounds great except you have to raise I to drive higher switching frequency leading to higher I2R losses.
So efficiency at device level is not going to show the complete picture.
Optimization of metal layers and PD network is more and more critical. Accurate parasitic extraction would be hell for these processes.
At SoC level things like interconnect efficiency plays big role in compute efficiency due to energy used during data movement

I don't know if this has been shared around here before, but these slides that some of AMD's engineers have put together reveal some of the challenges they faced during the design of Zen 2 on N7. I'd imagine since Zen 4 is also on a node shrink, there was many similar, if not larger challenges moving to N5.

Zen 2: The AMD 7nm Energy-Efficient High-Performance x86-64 Microprocessor Core

Zen 2: The AMD 7nm Energy-Efficient High-Performance x86-64 Microprocessor Core - Download as a PDF or view online for free

www.slideshare.net

 

[moinmoin]

I don't know if this has been shared around here before, but these slides that some of AMD's engineers have put together reveal some of the challenges they faced during the design of Zen 2 on N7. I'd imagine since Zen 4 is also on a node shrink, there was many similar, if not larger challenges moving to N5.

Zen 2: The AMD 7nm Energy-Efficient High-Performance x86-64 Microprocessor Core

Zen 2: The AMD 7nm Energy-Efficient High-Performance x86-64 Microprocessor Core - Download as a PDF or view online for free

www.slideshare.net

We had a thread nearly two years ago about several of those slides, but I wasn't aware of the whole set, thanks.

AMD's efforts involved in moving to TSMC's N7, advantages for going with chiplets (warning: many images)

I think this has been posted in four threads so far or something like that. Let's talk about these interesting slides in this dedicated thread instead. @rickxross mentioned this Japanese article with plenty English slides about AMD's efforts involved in moving to TSMC's N7 7nm node. @uzzi38...

forums.anandtech.com

 

[eek2121]

Imagine if yields were good enough for Epyc between the regular and the F to the point where AMD cancels Raphael.

AMD will never do that. They would be giving up a 300-billion-dollar rapid growth market. AMD is doing what they can with the constraints they have. Also, limiting themselves to a single market puts them at risk if they lose leadership in that market thanks to a changing competitive environment.

 

[BorisTheBlade82]

From a pure short-term revenue perspective that would be quite plausible - even more so WRT the consumer GPU market. Pure $/7nm mm2 is rather laughable compared to EPYC.

I think the reasons that this is a no-go are a mixture of brand visibility, trust of the markets in them and long-term-thinking.

 

[Ajay]

High speed IO makes sense to use some EM simulators on the routing. T-coil's especially need to be EM simulated as they rely on the mutual inductance between two spiral inductors.

So, two things. There aren't going to be any spiral inductors on chip - so I'm scratching my head on that one.
Secondly, The EM modelling tools were used for the TSVs. I'm not sure what the total stack height is for the TSV, but even if it's 200 μm, still short, but, apparently exhibiting the properties of a wave guide.

Of course, my E&M and Analog/Digital electronics courses were completed, uh, over 30 years ago - so there is that. Too bad I lost my TA position just b/4 starting my masters in EE (budget cuts at the university ). I wanted to become an ASIC designer.

 

[Mopetar]

Imagine if yields were good enough for Epyc between the regular and the F to the point where AMD cancels Raphael.

That's unlikely as not every chiplet is fit for use in server products. Thers's always going to be a bin a functional parts that just can't hit the TDP requirements that AMD sets. Not having a monolithic die would always enable some retail consumer parts to exist in order to maximize revenue by selling chiplets they otherwise couldn't.

Yeah, just need a market share above a certain point to make enough profit and then product positioning to get better GMs

I don't think that's true. Look at Apple who had a small share of the overall market, but the highest profits. You make the most profit by targeting the most profitable markets and market segments.

It's usually the companies with the largest market shares that are among the least profitable. You might have 30% of the market, but if it's the bottom 30% then it may not even be half as profitable as the top 3%.

 

[DisEnchantment]

So, two things. There aren't going to be any spiral inductors on chip - so I'm scratching my head on that one.

There are. The GMI2 IFOP PHY has on die T-Coils.

Secondly, The EM modelling tools were used for the TSVs. I'm not sure what the total stack height is for the TSV, but even if it's 200 μm, still short, but, apparently exhibiting the properties of a wave guide.

They would need to model the integrated inductor device made from TSVs too, if at all they plan to make that patented IVR die. This stacked IVR will be kind of the AMD equivalent of the Intel DLVR.
They have a conductor with TSVs and RDL traces surrounding it to which have a small EM field.

 

[Ajay]

There are. The GMI2 IFOP PHY has on die T-Coils.

Ugh. Forgot about high speed IO, even though @Hitman928 mentioned it

They would need to model the integrated inductor device made from TSVs too, if at all they plan to make that patented IVR die. This stacked IVR will be kind of the AMD equivalent of the Intel DLVR.
They have a conductor with TSVs and RDL traces surrounding it to which have a small EM field.

AH, forgot that LinkedIn post was on forward looking technologies (GMI3 listed, which points to Zen4).

Pardon me whilst I go look for some crow to eat. <<Hangs head in shame>>

 

[DisEnchantment]

Made a quick estimation of the probable efficiency gains using AMD's announced process characteristics for Zen4 and Zen 2 N7 data from the ISSCC presentation. Data is for 4C/8T.
N5 does certainly make those efficiency look sweet up until 4 GHz