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!

 

[nicalandia]

That's a lot of power being drawn by the IO dies.

Only 360 Watts for the 128C part(as per YuuKi_AnS)

 

[Timmah!]

Do you think that all core boost speed on the 3D cache models will be significantly slower than the regular versions?

This is the million dollar question. If the 7800 top boost is 400MHz lower than 7700 and is limited to 5GHz, will it clock to this value on all cores (like 7700x does i presume) or will it clock 300~400 MHz lower in that situation as well? Does 5800x3D clock lower in all core scenario compared to its limited single core boost?

 

[PJVol]

AMD Bergamo 10 Chiplet layout.

Is it actually two physical IODs? GMI numbering seem to suggest 12 ccds ?

 

[nicalandia]

Is it actually two physical IODs? GMI numbering seem to suggest 12 ccds ?

Two individual IODs

 

[Saylick]

Two individual IODs

Sounds like they now have a total of three IODs in production, up from two in the last generation.

I presume this third, new IOD is the same one that's used in Siena.

 

[moinmoin]

AMD Bergamo 10 Chiplet layout.

View attachment 74083

That chart leads to more questions than it answers. No connection between the two IODs? What's up with the odd GMI numbering? What about all the other I/O like memory and PCIe lanes?

I presume this third, new IOD is the same one that's used in Siena.

That'd be a neat solution.

 

[Timorous]

So each CCD has 2x 8c 16MB CCXs. Be interesting to see the transistor density but they look like phoenix CCXs at a high level.

 

[blckgrffn]

I understand some gnashing of teeth here but the vcache is marketed for games and sits on a 8C/16T CCD. Not that many cache dependent games (that tend to lean on main thread or two) are going to escape from there.

With the full blown tiny core/big core approach from Intel for what a year plus now I fully expect schedulers to be able to figure this out. As others have pointed out, crowd the CPU the cache until its full is the most straight forward approach. The corner cases where you'd like the extra boost vs the cache are darn niche. I get that we are pedantic here but, again, this is for gaming first and foremost.

If I was buying a first gen 3D cache part on AM5 for gaming I wouldn't even consider the wider parts. No worries about anything, no paying for cores I would super rarely use. The actual frequency uplift from the 5800X3D seems really solid as well.

If you are using all the cores all the time, it'll be slightly more heterogenous, but still not quite as much as the current Intel approach.

What I want is benchmarks

 

[Joe NYC]

Has it been revealed if the V-Cache is on N6, as has been rumored? Or is it N5?

 

[Hitman928]

AMD Bergamo 10 Chiplet layout.

View attachment 74083

What is the source for this?

 

[In2Photos]

They are awkward in the sense that they offer a selling point for which to make use of no OS and software is prepared for yet.

I seriously doubt that we are the first ones to realize that this may need some changes on the OS. Surely AMD realizes this which is why AMD said they have been working with Microsoft already.

I understand some gnashing of teeth here but the vcache is marketed for games and sits on a 8C/16T CCD. Not that many cache dependent games (that tend to lean on main thread or two) are going to escape from there.

With the full blown tiny core/big core approach from Intel for what a year plus now I fully expect schedulers to be able to figure this out. As others have pointed out, crowd the CPU the cache until its full is the most straight forward approach. The corner cases where you'd like the extra boost vs the cache are darn niche. I get that we are pedantic here but, again, this is for gaming first and foremost.

If I was buying a first gen 3D cache part on AM5 for gaming I wouldn't even consider the wider parts. No worries about anything, no paying for cores I would super rarely use. The actual frequency uplift from the 5800X3D seems really solid as well.

If you are using all the cores all the time, it'll be slightly more heterogenous, but still not quite as much as the current Intel approach.

What I want is benchmarks

Exactly!

 

[nicalandia]

What is the source for this?

Information about the 10 Die AMD EPYC 9754 Bergamo.

10 Dies per CPU on the EPYC 9754 Bergamo


360 Watts for the Production Release sample 9754 Bergamo. That's 2.88 Watt per Core. Mighty Impressive

 

[Kaluan]

I don't think they are awkward, instead I feel like they provided the consumer with enough skus for you to make a choice based on the following scenarios.

Gaming only - 7800X3D
Productivity only - 7900X or 7950X
Mixed use - 7900X3D or 7950X3D

The only pushback seems to be about the core clocks being dialed back and lack of vcache on both CCDs. But where does a 7900X3D or 7950X3D with the same core clocks as the non X3D versions and double vcache really shine? What type of workloads? Seems like it would be very few use cases.

Also may I just add, with all this negativity here, I don't think I've seen one(1) person comment (let alone positively) that we're getting 7000X3D in February, not the rumored March or later. That's a good thing? IDK I'm not sure anymore with all the gloom tinting. Everyone's too focused pointing and debating shortcomings, real or imagined or just minutiae.

Neither has there been much talk about expected pricing. I would be happy with $399 MSRP for the 7800X3D, slotting in above $349 quasi-MSRP 7700X and $329 MSRP (boxed w/ a cooler) 7700. But it's probably gonna be $449 or more.

 

[Saylick]

Has it been revealed if the V-Cache is on N6, as has been rumored? Or is it N5?

No confirmation that I've seen. Just guesses.

If I were a betting man, it's likely N6. SRAM doesn't scale with newer nodes, so why spend the money fabbing it on N5? Seeing as how the voltage characteristics of the V-cache modestly improved, that's likely just the uplift from N7 to N6.

 

[Hitman928]

Information about the 10 Die AMD EPYC 9754 Bergamo.

10 Dies per CPU
View attachment 74086

360 Watts for the Production Release sample 9754 Bergamo. That's 2.88 Watt per Core. Mighty Impressive
View attachment 74087

So it’s someone’s photoshop creation based on that bios info?

 

[nicalandia]

So it’s someone’s photoshop creation based on that bios info?

Yes,




The two IODs will likely look like a single one on package. That's a lot of routing to do.

 

[In2Photos]

Also may I just add, with all this negativity here, I don't think I've seen one(1) person comment (let alone positively) that we're getting 7000X3D in February, not the rumored March or later. That's a good thing? IDK I'm not sure anymore with all the gloom tinting. Everyone's too focused pointing and debating shortcomings, real or imagined or just minutiae.

Neither has there been much talk about expected pricing. I would be happy with $399 MSRP for the 7800X3D, slotting in above $349 quasi-MSRP 7700X and $329 MSRP (boxed w/ a cooler) 7700. But it's probably gonna be $449 or more.

I've been waiting a while to do a new build for myself so I'm happy with the February release. My guess is that it will be the end of February and not the beginning which would be very close to the original March rumor. I have a feeling $449 is pretty accurate but $399 or $429 are also possible.

 

[Kaluan]

This is the million dollar question. If the 7800 top boost is 400MHz lower than 7700 and is limited to 5GHz, will it clock to this value on all cores (like 7700x does i presume) or will it clock 300~400 MHz lower in that situation as well? Does 5800x3D clock lower in all core scenario compared to its limited single core boost?

5800X3D clocks to ~4,32GHz in all core loads (like rendering) and 4,45GHz in single core. So not a lot of difference. It also draws less power than 5800X under similar heavy loads.
All of this just points to me that it's possibly artificially limited.

Curve Optimizer or PBO may claw back some of that clockspeed or it may not do much. But it's certainly something we should hope reviewers dig into properly (SkatterBench where u at bro?).

Fmax is also a question, 7700X actually goes up to 5,55GHz on multiple cores in lightly threaded or light multithreaded loads. Real world clocks of stock 7800X3D may be different than advertised 5GHz as well.

 

[biostud]

Also may I just add, with all this negativity here, I don't think I've seen one(1) person comment (let alone positively) that we're getting 7000X3D in February, not the rumored March or later. That's a good thing? IDK I'm not sure anymore with all the gloom tinting. Everyone's too focused pointing and debating shortcomings, real or imagined or just minutiae.

Neither has there been much talk about expected pricing. I would be happy with $399 MSRP for the 7800X3D, slotting in above $349 quasi-MSRP 7700X and $329 MSRP (boxed w/ a cooler) 7700. But it's probably gonna be $449 or more.

Could be February 28th, so....

 

[Kaluan]

Could be February 28th, so....

Yeah lol

Still, if they said March (and ment March 1st), like the rumor, everyone would just assume March 31st, so still an improvement 😅

 

[CP5670]

I'm thinking of getting one of these and an X670E setup. The 7800X3D and 7900X3D have pretty different boost clocks, I wonder if it will be worth getting the higher model even if you don't need the cores.

 

[biostud]

I'm thinking of getting one of these and an X670E setup. The 7800X3D and 7900X3D have pretty different boost clocks, I wonder if it will be worth getting the higher model even if you don't need the cores.

Most believe the boost clocks only will be the CCD without cache, while the one with cache will probably top out at 5Ghz like the 7800X3D.

 

[Hitman928]

Yes,

View attachment 74088


The two IODs will likely look like a single one on package. That's a lot of routing to do.

How will they look like 1 IOD if they are separate dies? How would the data transfer between the groupings be handled? If this is user created, it should be made clear as such, rather than attempting to make it look like an AMD slide.

 

[moinmoin]

I seriously doubt that we are the first ones to realize that this may need some changes on the OS. Surely AMD realizes this which is why AMD said they have been working with Microsoft already.

That's the least we should expect. But AMD really ought to work on being able to offer required software and OS tweaks already at the time when their products launch. "Fine wine" makes for a nice meme but essentially expanding it across the product range (now potentially with "Ryzen AI" as well etc.) is not exactly a good look.

 

[Hail The Brain Slug]

Most believe the boost clocks only will be the CCD without cache, while the one with cache will probably top out at 5Ghz like the 7800X3D.

AFAIK all of this this was confirmed independently by multiple press members contacting AMD.

High boost on non cache CCD

5.0ghz/1.4v max on cache CCD

Albeit PBO and CO will be enabled this time around, so potentially able to get >5.0GHz on the cache CCD granted thermals and sample quality are good