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!

 

[jpiniero]

It's also cheaper per wafer. N6 requires less steps on some of the machines that are a production bottleneck, they can push out substantially more N6 wafers out of the same plant.

Cheaper for TSMC possibly but there's no real reason to charge less for wafers when demand is unlimited to the point where they are hiking prices. That's the point I'm trying to get across, things have been blown sky high.

 

[soresu]

Cheaper for TSMC possibly but there's no real reason to charge less for wafers when demand is unlimited to the point where they are hiking prices. That's the point I'm trying to get across, things have been blown sky high.

It should be more or less the same for fabs and their customers - it's mostly their customers customers (ie consumers and bulk buyers) downstream that suffer.

The demand has usually been high since smartphones exploded, which is why Apple pour so much money into the problem to keep a giant share of cutting edge fab capacity to themselves.

At the end of the day TSMC raising prices only raises the barrier of entry for new semicon players and consequentially reduces their possible chances of profit once Apple have their fill of a node.

That being said raising those prices does help them to build out further capacity, so I can see its short term benefits.

 

[scannall]

Cheaper for TSMC possibly but there's no real reason to charge less for wafers when demand is unlimited to the point where they are hiking prices. That's the point I'm trying to get across, things have been blown sky high.

If N6 costs TSMC say %25 less per wafer (Not sure of actual numbers, just using fillers to explain it.) then charging %10 less would be enough to move some production to that, and still end up with a net gain.

 

[maddie]

Cheaper per transistor, not cheaper per wafer. And there's no guarantee that TSMC would pass on any cost savings on now, given the price hikes.

Now I know you're just making stuff up.

TSMC specifically stated that you can migrate from N7 to N6 with minimal work and it'll be cheaper. This means you do not get the density increase, which is equal to saying cost/area is less.

You are fabricating an alternate reality here, by claiming cost/transistor but not cost/area is lest.

 

[jpiniero]

TSMC specifically stated that you can migrate from N7 to N6 with minimal work and it'll be cheaper. This means you do not get the density increase, which is equal to saying cost/area is less.

Was that before or after the price hikes?

 

[maddie]

Was that before or after the price hikes?

Are you arguing that internally to TSMC we are now having cost increases between N7 & N6. This is distinct from the general external factors raising chip prices worldwide, eg substrate, increased demand, etc.

External factors affect all lines. The auto makers don't use latest tech and are affected like everyone else.

In other words external factor price hikes do not change the relative cost positions of nodes. Internal factors do.

If external, then N6 < N7 remains as before.

 

[jpiniero]

Are you arguing that internally to TSMC we are now having cost increases between N7 & N6.

I don't believe TSMC was ever charging less per wafer for N6 compared to N7. But if any customer got any deals, those are dead. It could still be cheaper per transistor and you get the benefit of additional supply.

 

[maddie]

I don't believe TSMC was ever charging less per wafer for N6 compared to N7. But if any customer got any deals, those are dead. It could still be cheaper per transistor and you get the benefit of additional supply.

Any hard info or just a belief?

 

[eek2121]

I don't believe TSMC was ever charging less per wafer for N6 compared to N7. But if any customer got any deals, those are dead. It could still be cheaper per transistor and you get the benefit of additional supply.

They are charging xx% (double-digit) less per wafer. Whether or not you believe it is irrelevant. It is happening.

 

[moinmoin]

I don't believe TSMC was ever charging less per wafer for N6 compared to N7.

Your belief is off then. TSMC always positioned N6 as the mainstream node compared to N7 being the high performance one. And N6 is mainstream for being cheaper.

 

[jpiniero]

Any hard info or just a belief?

Only thing I can say for sure is that N7 and N5 cost more than they did 2 years ago. No reason to think N6 is any different and any deals are dead. Normally nodes get cheaper over time... not anymore.

Your belief is off then. TSMC always positioned N6 as the mainstream node compared to N7 being the high performance one. And N6 is mainstream for being cheaper.

That was before the price hikes. N6 was to me always talked about the better output and not being cheaper in the first place.

 

[Hougy]

I don't know if this is an actual picture or if someone made a 3D render/print with known information, but it checks out in every aspect.

After The SP5 Socket, AMD's EPYC Genoa CPU With 12 Next-Gen Zen 4 CCDs on 5nm Node With Up To 96 Cores Pictured

AMD's 5nm Zen 4 powered EPYC Genoa server CPU has been pictured in its full-fat 12 CCD configuration for the SP5 server platform.

wccftech.com

View attachment 59750

Why are the dies so thick? Will it support cache stacking near launch?

 

[jpiniero]

Why are the dies so thick? Will it support cache stacking near launch?

No

 

[moinmoin]

That was before the price hikes. N6 was to me always talked about the better output and not being cheaper in the first place.

Even before the price hikes N6 was the cost optimized compatible follow on to N7.

 

[nicalandia]

Why are the dies so thick? Will it support cache stacking near launch?

No, thats why it seems to be fake

 

[igor_kavinski]

No, thats why it seems to be fake

And if it is not fake, what does that mean?

 

[eek2121]

Only thing I can say for sure is that N7 and N5 cost more than they did 2 years ago. No reason to think N6 is any different and any deals are dead. Normally nodes get cheaper over time... not anymore.



That was before the price hikes. N6 was to me always talked about the better output and not being cheaper in the first place.

My information is accurate as of the beginning of Q2 2022.

 

[DrMrLordX]

AMD can absorb it fine, they just have to raise prices.

They will definitely raise prices. After a fashion.

As a segue from that point: where does everyone think the Raphael lineup on AM5 will start as a baseline? What will be the low-end part? We had this discussion going in the old Zen3/Ryzen 4000 discussion thread where technically it was off-topic, so I'm trying to move it here if there's anything left to be discussed.

Personally I'm looking at an 8c part in the $350-$399 range as the base Raphael. The stack will move up to . . . whatever winds up being the halo 170W TDP part for AM5 at a launch MSRP higher than the 5950X. Whether that's 24c or not remains to be seen. In any case, I don't see a 6c successor launching on AM5 yet. We might see one down the road as a budget part once Raphael has been on the market for awhile.

 

[igor_kavinski]

No, thats why it seems to be fake

96-Core Processor Made of Chiplets

CEA-Leti bound dozens of processor cores together by breaking them up and stitching them back together on a separate piece of silicon

spectrum.ieee.org

That's thick too and doesn't look fake.

 

[RnR_au]

In any case, I don't see a 6c successor launching on AM5 yet. We might see one down the road as a budget part once Raphael has been on the market for awhile.

idk... surely AMD must be getting some bad dies on occasion, despite TSMC's godly yield.

 

[jpiniero]

Personally I'm looking at an 8c part in the $350-$399 range as the base Raphael. The stack will move up to . . . whatever winds up being the halo 170W TDP part for AM5 at a launch MSRP higher than the 5950X. Whether that's 24c or not remains to be seen. In any case, I don't see a 6c successor launching on AM5 yet. We might see one down the road as a budget part once Raphael has been on the market for awhile.

Remember the core count per chiplet hasn't changed. The only thing that's changed is that Epyc gets 3 per group and that's unlikely to happen on Ryzen. The 24 core almost certainly involves leaky Bergamo, and that won't be ready at launch.

What I think is going to happen at launch is that there won't be a Ryzen 5; only 7 and one 9. I could be talked into the Ryzen 7 being 12 cores instead of 8 if they feel like they need that. The Ryzen 9 would then be 12 or 16. The 5950X replacement would be the Bergamo part.

 

[uzzi38]

idk... surely AMD must be getting some bad dies on occasion, despite TSMC's godly yield.

N5 is higher yielding than N7. Whatever bad dies they do you can be certain are getting thrown at EPYC first if anywhere, in particular for the eventual F SKUs.

 

[DrMrLordX]

The 5950X replacement would be the Bergamo part.

The 3950X wasn't available at Matisse's launch. AMD may do the same for Raphael's true halo part.

 

[Mopetar]

Cheaper for TSMC possibly but there's no real reason to charge less for wafers when demand is unlimited to the point where they are hiking prices. That's the point I'm trying to get across, things have been blown sky high.

Because price is relative, but production capacity is fixed regardless of what they charge.

Suppose that N7 wafers cost some amount of money and that X quantity can be produced in some amount of time. Now suppose that TSMC decides to charge 10% less for N6, but can produce 1.2X wafers in that same time period. If all N7 production were moved to N6 then TSMC would make 8% more money, even though they're changing less per wafer.

Obviously there are numbers you could pick where TSMC only breaks even or even loses money, but they don't reduce prices to that extent.

 

[jpiniero]

Suppose that N7 wafers cost some amount of money and that X quantity can be produced in some amount of time. Now suppose that TSMC decides to charge 10% less for N6, but can produce 1.2X wafers in that same time period. If all N7 production were moved to N6 then TSMC would make 8% more money, even though they're changing less per wafer.

Those EUV machines aren't cheap. Even before the hikes they can justify keeping pricing the same or more because of the density increase.

AMD will be fine as long as they can pass on the extra costs.