Speculation: Ryzen 4000 series/Zen 3

[inf64]

So let me lay out what I think the competitive desktop CPU landscape will be for Q4 2020-Q4 2021. Intel is beyond sc***ed with their Rocket Lake Vs even Zen2 parts. Zen3 will bring higher (than Willow Cove) IPC, higher clocks, much better power draw/thermals and 2x the core count. The only reason why anyone should be buying a Rocket Lake platform is if it's much cheaper than Zen2/Zen3 - this will not happen. APU segment will also be brutally bad for intel, Cezzane will annihilate Tiger Lake and AMD will have Navi23 in Cezzane based laptops, meaning Nvidia is left behind as well.

MAYBE in 2022 intel could become semi-competitive IF whatever Cove derivative they have out at that time can match the clock speeds and core core counts of Zen4 (which could happen). Until then, I think it will be a curb stomping by AMD.

 

[DrMrLordX]

Cezzane will annihilate Tiger Lake

Depending on availability, Cezanne may face Alder Lake-P for most of its time on the market. Also I am not 100% sure that AMD is going to switch to Navi for their APUs. AMD isn't standing still on CDNA though . . . so we may see updated Vega30-ish graphics in Cezanne.

 

[uzzi38]

Depending on availability, Cezanne may face Alder Lake-P for most of its time on the market. Also I am not 100% sure that AMD is going to switch to Navi for their APUs. AMD isn't standing still on CDNA though . . . so we may see updated Vega30-ish graphics in Cezanne.

Yeah, we'll have to see on Cezanne. As for Alder Lake-P, well I honestly wonder what that'll be. I mean, I'm guessing big.LITTLE like desktop, but 4+4? 4+8? 8+4?

Last doesn't really make sense product wise - cutting down on the little cores while leaving all 8 big cores intact, and the first could be a perf degredation over 8c TGL-H, so one would assume 4+8 then? In which case I'd argue Cezanne still will likely have a strong position in the market...

 

[DrMrLordX]

@uzzi38

Well like another posted reminded me in the Intel uarch thread, Intel to date has mirrored core counts exactly between -S and -H dice. So at the top-end, I expect an 8+8 Alder Lake-P @ 45W TDP. Whether or not any of the lower-power variants mirror that core count remains to be seen.

edit: it was @jpiniero

 

[french toast]

My guess is a chiplet like approach with some big little tech.
4x4 chiplets, combining 2 chiplets with 4 big willow cove/golden cove? and new Atom cores on 10nm+++.
Although I have another thought that we will see full 8 big core chiplets and 8 little cores chiplets, then Alderlake S with 16 full golden cove cores SKU on 10+++.
Could be very interesting mid 2021.

Saying that we expect N5P zen 4 at Q3/Q4 2020..

 

[uzzi38]

@uzzi38

Well like another posted reminded me in the Intel uarch thread, Intel to date has mirrored core counts exactly between -S and -H dice. So at the top-end, I expect an 8+8 Alder Lake-P @ 45W TDP. Whether or not any of the lower-power variants mirror that core count remains to be seen.

edit: it was @jpiniero

The point of -P is to make -U and -H share dies, so -H no longer shares a die with -S tho.

 

[CHADBOGA]

My guess is a chiplet like approach with some big little tech.
4x4 chiplets, combining 2 chiplets with 4 big willow cove/golden cove? and new Atom cores on 10nm+++.
Although I have another thought that we will see full 8 big core chiplets and 8 little cores chiplets, then Alderlake S with 16 full golden cove cores SKU on 10+++.
Could be very interesting mid 2021.

Saying that we expect N5P zen 4 at Q3/Q4 2020..

I seriously doubt we will see Zen 4 till Q1/Q2 2021

 

[DrMrLordX]

The point of -P is to make -U and -H share dies, so -H no longer shares a die with -S tho.

It may be that -P and -S will share dice.

I seriously doubt we will see Zen 4 till Q1/Q2 2021

Genoa might show up then (Q1 2021) as early silicon, but commercial Zen4 may not appear until 2022. Bleh!

 

[randomhero]

Depending on availability, Cezanne may face Alder Lake-P for most of its time on the market. Also I am not 100% sure that AMD is going to switch to Navi for their APUs. AMD isn't standing still on CDNA though . . . so we may see updated Vega30-ish graphics in Cezanne.

I respectfully disagree.
By all means rdna2 is ready since its confirmed for consoles and graphics cards. So next gen apu should almost certainly have rdna2 as gpu part.

 

[soresu]

AMD isn't standing still on CDNA though . . . so we may see updated Vega30-ish graphics in Cezanne.

I would not expect anything of CDNA in consumer products going forward - the whole point of CDNA is to separate the workstation/server/datacenter compute markets that stray from typical graphics use cases, with RDNA for everything and anything else in the consumer and graphics/video/light compute oriented workstation use cases.

I think we might well see a limited CDNA spill over in future RDNA generations - ie if CDNA has 4 tensor optimised units per CU, we might get 1 tensor unit in an RDNA3+ CU.

After all, machine learning still has significant potential for computer graphics, and it would be dangerous to allow nVidia to gain the high ground by default in this area - especially as ML is ideal for denoising low sample per pixel frames for RT graphics.

There are also papers that already exist for improving RT sampling through ML, this may allow for more samples or bounces per pixel at the same frame rate in real time.

It's not at all impossible that we may see a CDNA based APU in the future for the enterprise market though, especially if they stack HBM on CDNA chiplets for adequate mem bandwidth.

 

[piokos]

4400g 4c (Renoir)
4400x 4c
4500g 6c (Renoir)
4500x 6c
4600g 8c (Renoir)
4600x 8c
4700x 10c
4800x 12c
4900x 14c
4950x 16c

I really doubt they sell so many items to make such a dense lineup sensible. IMO it'll stay at 4-6-8-12-16 cores.
Also, the difference between 14C and 16C would be minimal, so it would take away some of the "premium feel" of the top model.

The big question IMO: would it be possible to put a weak IGP onto the I/O die? Or will they change the layout for Zen4?
Because for now APUs are limited to 8 cores. Intel will offer 10 cores this year (still on 14nm) and who-knows-how-many on 10nm (they could easily fit 16 Sunny Cove cores on LGA1200).
AMD hasn't really got into OEM business desktops yet - and it will get ever harder from now on.

 

[Veradun]

I really doubt they sell so many items to make such a dense lineup sensible. IMO it'll stay at 4-6-8-12-16 cores.

I doubt as well they will do that, but for science I wanted to put up a lineup that makes sense using those layouts.

Keep also in mind that when you remove one from each couple 3600/X and 3700X/3800X you already made the room required for 10c and 14c :>

Current core creep is making less cores less appealing and diversification through cores instead of X/non-X might make sense.

 

[moinmoin]

Saying that we expect N5P zen 4 at Q3/Q4 2020..

I seriously doubt we will see Zen 4 till Q1/Q2 2021

Genoa might show up then (Q1 2021) as early silicon, but commercial Zen4 may not appear until 2022. Bleh!

Or will they change the layout for Zen4?

What are y'all talking about Zen 4 with Zen 3 coming next first (and this thread being about Zen 3)?

 

[randomhero]

@Veradun posted that list based on my speculation as an example.
That heavy segmentation in product lineup is very unlikely but not improbable.

 

[maddie]

Intel only went that way to extract as much margin as possible as they were already very dominant and performance progress was fairly slow. When you're sprinting, all those incremental SKUs only complicate matters from inventory control to production cycle control, to even as simple as wasted hours arguing on the best mix.

 

[LightningZ71]

I was under the impression that Intel was sacrificing the iGPU section on their ten core consumer products, i.e., they are going to have two different runs of consumer i series dies, one with an iGPU that tops out at 8 cores, and feeds their non-F series products, with failed iGPU die recovery going to low bin -F products, and another without that tops out at 10 cores, and feeds their -F series products. Given their size and volume, its not that cost prohibitive to do this, as those products will be on different lines anyway, the dies are HIGHLY related, and Intel has many different lines active. This could also allow MINOR process tweaks to the lines making the 10 core dies that would be problematic for the iGPU section, but benefit the CPU cores. I've seen a lot of speculation back and forth on that.

(the above was all wrong, crow eaten)

As for ZEN3 and it's consumer products, I don't think that there will be a lot of changes in the IO die for the 4000 series non-iGPU desktop products, aside from a possible process change. I don't see AMD adding an iGPU to the I/O die until they shift to socket AM5. It doesn't make a lot of sense for a high end product to use an iGPU on AM4 as it has to sacrifice some CPU served PCI lanes to multiplex out the video signals. Its a market limitation that's going to serve to limit it. Desktop Renoir will be plenty for over 90% of the market, and it's follow on will surely be fine there as well.

EDIT: I'm officially eating crow, they have non -F 10 core products.

 

[piokos]

Keep also in mind that when you remove one from each couple 3600/X and 3700X/3800X you already made the room required for 10c and 14c :>

On SKU count, maybe. But I meant the lineup differentiation.
You don't want CPUs landing too close to each other. 16C could be 14% faster than 14C, but realistically - because of power limits - this would probably be in the 5-10% range.

Current core creep is making less cores less appealing and diversification through cores instead of X/non-X might make sense.

Absolutely. That -X naming was a pretty bad idea from the start and it's meaningless.

Let's hope "X" goes away completely.

 

[piokos]

What are y'all talking about Zen 4 with Zen 3 coming next first (and this thread being about Zen 3)?

Well, I asked about a possibility to free a CPU slot on the APU - if it's possible with this or next gen.

Frankly, Zen3 is where Intel buyers are every 2 years. It's the last gen on socket and - honestly - who thinks Zen2 doesn't have enough juice?
So we end up with a generation that adds no features, not much performance and doesn't open an upgrade path.

This makes me wonder what AMD will do with Zen2 chips. Will they keep them on offer or not? Will the price drop like it did with Zen+?

 

[maddie]

Well, I asked about a possibility to free a CPU slot on the APU - if it's possible with this or next gen.

Frankly, Zen3 is where Intel buyers are every 2 years. It's the last gen on socket and - honestly - who thinks Zen2 doesn't have enough juice?
So we end up with a generation that adds no features, not much performance and doesn't open an upgrade path.

This makes me wonder what AMD will do with Zen2 chips. Will they keep them on offer or not? Will the price drop like it did with Zen+?

Want to expand?

 

[piokos]

Want to expand?

We know there will be some performance increase vs Zen2. Nothing else changes, right?
Still PCIe4.0, still DDR4, most likely the same instruction set and other features.

In other words: We'll have a very dense choice of Zen2 and Zen3 CPUs if AMD decides to keep both in production.
On the top there will be some "4950X" with unprecedented performance.
But down the line you'll always have a choice between a Zen3 CPU with higher single-core performance and older Zen2 CPU with more cores for the money.

 

[moinmoin]

We know there will be some performance increase vs Zen2. Nothing else changes, right?

I thought we were told it's a whole new family?

 

[awesomedeluxe]

I respectfully disagree.
By all means rdna2 is ready since its confirmed for consoles and graphics cards. So next gen apu should almost certainly have rdna2 as gpu part.

AMD is definitely considering sticking with Vega for APUs. RDNA(2) has caused headaches for them. And the demand for powerful internal graphics is unclear.

I can tell you with certainty they do not have a decision right now. This could be a lategame call.

 

[moinmoin]

RDNA(2) has caused headaches for them.

Can you expand on that?

 

[maddie]

AMD is definitely considering sticking with Vega for APUs. RDNA(2) has caused headaches for them. And the demand for powerful internal graphics is unclear.

I can tell you with certainty they do not have a decision right now. This could be a lategame call.

Headaches? In what sense?

Will Vega as a distinct line still exist? GCN has split into CDNA & RDNA going forward.

 

[teejee]

AMD is definitely considering sticking with Vega for APUs. RDNA(2) has caused headaches for them. And the demand for powerful internal graphics is unclear.

I can tell you with certainty they do not have a decision right now. This could be a lategame call.

Of course they have decided that, they probably have early samples available, and regardless no chance that such major architectural decision isn't decided yet for Zen 3 APU.