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!

 

[Joe NYC]

Absolutely it is. As many have mentioned, AM4 + Zen 3 + Vcache + DDR4 is a solid mature platform for those who don't want to spend extra on the latest and greatest while offering competitive performance. AM5 + Zen 4 + DDR5 would be the premium option for those who want the cutting edge and that extra oomph.

It's almost like Vcache is the CPU version of the design philosophy behind RDNA2, where instead of going with a wider memory bus and newer memory standard, they slapped on even more cache.

Yes, and also process technology where AMD has huge amount of capacity, where AMD can produce million and millions of CPUs.

Zen 4 on the new TSMC N5 process technology will be so capacity limited, that initially, we will probably be stuck with high scalper prices like first 6 month of desktop Zen 3.

 

[Joe NYC]

At least he isn't saying that AMD should use SMT4

That's probably what "5950x gamers" think they need.

 

[Thunder 57]

I think Threadripper as a gaming platform is not a good idea. The main theme behind Threadripper is "Thread Ripper", meaning MT performance, which is the opposite end of the spectrum of gaming CPU. And cost of the platform makes it a non-starter.

It seems that every CPU AMD is offering right now has an element that keeps it from being a fully optimized for gaming:

5800x:
+ closest to ideal, acceptable price
- not using the best binning dies
- IOD-CCD connection not ideal for gaming, limits overclocking, limits maxing out the memory

5950:
+ highest binning dies
- not fully maxed out because of power limit on 16 cores
- high cost
- wasted 2nd CCD
- extra memory coherency between CCDs
- IOD-CCD link see above

Cesanne:
+ good price
+ no IOD-CCD
+ good for overclocking and maxing out memory
- 65W limit
- 16 MB L3

Threadripper:
- everything

It would be interesting if AMD could add V-Cache to Cezanne, tune it it for high clock speed and 105W.

2nd choice would be to max out 5800x with highest bin dies and max V-Cache.

I know it is so frustratingly difficult for so many to understand, but AMD designs CPU's for general purpose abilities. AMD is not out there to build a "gaming CPU". If you want one of those, get a console. AMD builds CPU's that are capable of incredible things, from supercomputers to servers. If you can happen to game on one, great!

For redacted sake, if all we did was build computer hardware for entertainment or gaming, rather than scientific research or trying to find answers to questions that actually matter, we'd be no better off then those chumps in the movie "Idiocracy". It's like saying "Who cares about learning about the universe? Give me Half Life 3!".

Profanity is not allowed in the tech sub-forums.

Iron Woode
Super Moderator

 

[Timorous]

I know it is so frustratingly difficult for so many to understand, but AMD designs CPU's for general purpose abilities. AMD is not out there to build a "gaming CPU". If you want one of those, get a console. AMD builds CPU's that are capable of incredible things, from supercomputers to servers. If you can happen to game on one, great!

For redacted sake, if all we did was build computer hardware for entertainment or gaming, rather than scientific research or trying to find answers to questions that actually matter, we'd be no better off then those chumps in the movie "Idiocracy". It's like saying "Who cares about learning about the universe? Give me Half Life 3!".

All I get from this post is Half Life 3 confirmed.

 

[Joe NYC]

I know it is so frustratingly difficult for so many to understand, but AMD designs CPU's for general purpose abilities. AMD is not out there to build a "gaming CPU". If you want one of those, get a console. AMD builds CPU's that are capable of incredible things, from supercomputers to servers. If you can happen to game on one, great!

For redacted sake, if all we did was build computer hardware for entertainment or gaming, rather than scientific research or trying to find answers to questions that actually matter, we'd be no better off then those chumps in the movie "Idiocracy". It's like saying "Who cares about learning about the universe? Give me Half Life 3!".

I get it. AMD has limited resources, and cannot have optimized product every use case scenario.

But that was 2020, when Zen 3 launched. We are in 2021, and the main competitor is releasing gaming optimized SKU as a launch SKU for its next gen CPU design.

As a response to that, I am suggesting that AMD assembles the Lego pieces to also launch a gaming optimized CPU - to compete with Alder Lake.

I don't think this is a far fetched proposition, to expect AMD to launch a Halo product, using optimum configuration of existing dies to achieve the highest performance.

 

[Thibsie]

I don't think this is a far fetched proposition, to expect AMD to launch a Halo product, using optimum configuration of existing dies to achieve the highest performance.

That's not what you're asking for.
You asking for a dedicated development just for games.

 

[soresu]

But that was 2020, when Zen 3 launched. We are in 2021, and the main competitor is releasing gaming optimized SKU as a launch SKU for its next gen CPU design.

AMD are doing extremely well now, but even with that they still have only a fraction of the operating capital of either Intel or nVidia to develop new products.

(not to mention that fab capacity is also an ongoing issue now affecting SKU volumes)

Their Zen3D/Zen3V platform will nicely boost gaming performance already, there is no need for any further platform.

Then once Zen4 is ready it will increase overall IPC once again, not that a whopping big load of cache won't benefit other compute tasks beyond gaming with Zen3V.

 

[eek2121]

That's probably what "5950x gamers" think they need.

Some of my work (development) likes high core counts. However, gaming DOES benefit from the faster clock speeds of the 5950x. Many of my cores can hit 5ghz, and in many games, 4.7-5ghz on a few cores is the norm. You might be inclined to state that you can overclock the other chips, well, you can so that with the 5950x as well. I don’t overclock because I need the stability.

I get it. AMD has limited resources, and cannot have optimized product every use case scenario.

But that was 2020, when Zen 3 launched. We are in 2021, and the main competitor is releasing gaming optimized SKU as a launch SKU for its next gen CPU design.

As a response to that, I am suggesting that AMD assembles the Lego pieces to also launch a gaming optimized CPU - to compete with Alder Lake.

I don't think this is a far fetched proposition, to expect AMD to launch a Halo product, using optimum configuration of existing dies to achieve the highest performance.

The competition is not launching a “gaming oriented SKU”.

 

[Joe NYC]

That's not what you're asking for.
You asking for a dedicated development just for games.

No, I said nothing of the kind.

I was brainstorming which product configuration would have the best performance. Of the 2 I mentioned:

5800x + highest binned dies + maxed out V-Cache: can be assembled with the Lego pieces AMD will have by the end of the year.

Cezanne + V-Cache: is not something possible presently, because most likely, Cezanne is not set up for that with the TSVs and all. But as a future CPU, this would be a killer product across all categories, from laptop to corporate desktop to SFF boxes to gaming.

Unless AMD optimizes the CCD-IOD connection in the future so that it is not holding back the CPUs from best gaming performance

 

[Joe NYC]

AMD are doing extremely well now, but even with that they still have only a fraction of the operating capital of either Intel or nVidia to develop new products.

(not to mention that fab capacity is also an ongoing issue now affecting SKU volumes)

Their Zen3D/Zen3V platform will nicely boost gaming performance already, there is no need for any further platform.

Then once Zen4 is ready it will increase overall IPC once again, not that a whopping big load of cache won't benefit other compute tasks beyond gaming with Zen3V.

I was not suggesting a new platform somehow magically appear in 3 months (although I hope the B2 stepping will have some incremental improvements).

All AMD needs to do is to configure an SKU with a single, highest binned Zen 3 die max out the V-Cache, and hope for the best against Alder Lake.

 

[gdansk]

They will make a gaming optimized CPU. It'll just have extra cores so they can justify the higher MSRP. I don't get why that's so frustrating.

 

[LightningZ71]

VCache would be absolutely killer in an APU. Just using a 64MB VCache tile stacked on something as simple as the existing Cezanne die, IF IT WAS SO PROVISIONED, would be a dramatic performance boost, even if it only served as a cache for the iGPU, or, if it just served as a memory controller cache. It would save power on unneeded memory requests, and it would drastically improve the effective memory performance for the whole package.

 

[Joe NYC]

Some of my work (development) likes high core counts. However, gaming DOES benefit from the faster clock speeds of the 5950x. Many of my cores can hit 5ghz, and in many games, 4.7-5ghz on a few cores is the norm. You might be inclined to state that you can overclock the other chips, well, you can so that with the 5950x as well. I don’t overclock because I need the stability.

That's actually what I was referring to as sub-optimal.

5950x is clearly a sub-optimal CPU for 9 out of 10 current games, 10 out of 10 of older games.

The reason it wins some is because AMD dedicated the highest binned parts to this CPU, and it has the highest boost clock.

5800x configuration with a single CCD is better suited for gaming, if it had the best binning parts and highest boost clock. Single CCD would have more of the power budget, and the performance is not slowed down by resolving memory coherency between 2 CCDs.

The competition is not launching a “gaming oriented SKU”.

Maybe you missed the news that Intel is launching only the gaming oriented K SKUs this year, with DDR5, postponing all of the mainstream SKUs for next year.

 

[Thunder 57]

That's actually what I was referring to as sub-optimal.

5950x is clearly a sub-optimal CPU for 9 out of 10 current games, 10 out of 10 of older games.

The reason it wins some is because AMD dedicated the highest binned parts to this CPU, and it has the highest boost clock.

5800x configuration with a single CCD is better suited for gaming, if it had the best binning parts and highest boost clock. Single CCD would have more of the power budget, and the performance is not slowed down by resolving memory coherency between 2 CCDs.



Maybe you missed the news that Intel is launching only the gaming oriented K SKUs this year, with DDR5, postponing all of the mainstream SKUs for next year.

I still wasn't aware that AMD was developing gaming chips. Pretty sure they want to produce general central processing units. AMD and Intel pretty much develop chips with a "server first" mentality, as that is the lucrative market.

 

[Timorous]

Zen 3d does not need to be faster than Alder Lake in gaming imo, it just needs to be better value like Zen 2 was. Further with the initial offering for Alder being DDR5 I expect the platform cost itself to be a lot more expensive.

 

[Joe NYC]

I still wasn't aware that AMD was developing gaming chips. Pretty sure they want to produce general central processing units.

No need to twist my words. AMD has dies that can be assembled to optimize for different use scenarios.

Assembling an SKU that is optimized for gaming does not equal "developing gaming chips"

 

[Ajay]

Zen 3d does not need to be faster than Alder Lake in gaming imo, it just needs to be better value like Zen 2 was. Further with the initial offering for Alder being DDR5 I expect the platform cost itself to be a lot more expensive.

That's would be an option, but AMD will be following Adler Lake with Zen3 + vCache in fairly short order (which may be more expensive than Intel's CPU). So, I think the platform cost may be similar. Interesting times though!

 

[Thunder 57]

No need to twist my words. AMD has dies that can be assembled to optimize for different use scenarios.

Assembling an SKU that is optimized for gaming does not equal "developing gaming chips"

I don't mean to twist your words. It just annoys me when people think gaming is the goal. I've seen numerous reviews of ~32 core CPU's and they do gaming benchmarks on them. Sure, it'll game, but at a price that is way to high while delivering less performance.

 

[Joe NYC]

Zen 3d does not need to be faster than Alder Lake in gaming imo, it just needs to be better value like Zen 2 was. Further with the initial offering for Alder being DDR5 I expect the platform cost itself to be a lot more expensive.

Not every Zen 3d SKU needs to beat Alder Lake. But ideally, AMD will have 1 (of several) SKUs that will beat Alder Lake.

Because if this happens, AMD can happily sell 7nm Zen 3D dies all year 2022 (against Alder Lake), of which AMD can produce millions, without having to depend on Zen 4 timing for desktop and 5nm capacity from TSMC.

 

[Joe NYC]

I don't mean to twist your words. It just annoys me when people think gaming is the goal. I've seen numerous reviews of ~32 core CPU's and they do gaming benchmarks on them. Sure, it'll game, but at a price that is way to high while delivering less performance.

Exactly. Good gaming CPU is one with excellent single thread performance that takes into account the price.

So 5950x has too much overhead in resources (unnecessary extra CCD) and price. So less than ideal.

But I have faith that AMD will released an optimized SKU this fall that will have high binned die paired with enough V-Cache to successfully counter Alder Lake, and hopefully, at a price that is within reach of gamers. (closer to $500 than $1,000)

 

[Thunder 57]

Exactly. Good gaming CPU is one with excellent single thread performance that takes into account the price.

So 5950x has too much overhead in resources (unnecessary extra CCD) and price. So less than ideal.

But I have faith that AMD will released an optimized SKU this fall that will have high binned die paired with enough V-Cache to successfully counter Alder Lake, and hopefully, at a price that is within reach of gamers. (closer to $500 than $1,000)

Like some sort of 5800X variant? Sounds reasonable.

 

[darkswordsman17]

I get it. AMD has limited resources, and cannot have optimized product every use case scenario.

But that was 2020, when Zen 3 launched. We are in 2021, and the main competitor is releasing gaming optimized SKU as a launch SKU for its next gen CPU design.

As a response to that, I am suggesting that AMD assembles the Lego pieces to also launch a gaming optimized CPU - to compete with Alder Lake.

I don't think this is a far fetched proposition, to expect AMD to launch a Halo product, using optimum configuration of existing dies to achieve the highest performance.

So why exactly are you arguing against the product that literally is that?

And your argument is that Intel is only rolling out the highest of their high end ones for their next platform (which will require an entire platform upgrade including likely quite expensive memory), but saying that AMD needs to counter with a cheaper halo product? Until we know the performance and price of Intel's stuff, I'm not sure I'd say much about what AMD needs to do to counter it.

 

[Joe NYC]

Like some sort of 5800X variant? Sounds reasonable.

That's what I am hoping for.

Between the B2 stepping, selecting higher binned dies (with higher boost clocks) and V-Cache - AMD has some "dials" to to adjust.

Since Intel is releasing a gaming SKU that with projected power draw between 125 Watts and upwards of 200 Watts peak, with single goal of capturing the gaming crown, I hope AMD will have a decent response.

 

[Joe NYC]

So why exactly are you arguing against the product that literally is that?

And your argument is that Intel is only rolling out the highest of their high end ones for their next platform (which will require an entire platform upgrade including likely quite expensive memory), but saying that AMD needs to counter with a cheaper halo product?

With equally binned part, 5950x would lose in majority of games to 5800x.

So the best bet to counter Intel gaming SKU is with optimized 5800x CPU.

I am not saying that AMD should not launch 5950 with V-Cache. By all means. But it is going to be a weaker gaming CPU compared to maxed out 5800x.

 

[Thibsie]

AMD based their whole range on the fact (among others) that the more higher in the range you go, the higher MT you get bit you don't lose ST (except 5950 if I'm not mistaken) or very little which means you do not regress in ST as a price to get better MT.
(Not taking APU into account obviously).

If you break this, you lose the incentive to go 12/16 cores CPU, which mean those with better margin which means less sales of high margin CPU. Why on earth would AMD do such a stupid move?
Just for PR of 'I'm still the best at games' ?
How shortsighted this would be.