Question Zen 6 Speculation Thread

[Kepler_L2]

The other thing that keeps coming to mind are the relatively modest gains for Zen6 over Zen6 that were forecast on that slide that was circulated last year. They were barely looking to get 10% it seemed. Going from N4P to N3P and including fin flex where it can help should be able to net a good 5+% clock speed or more while also improving the power characteristics which would allow additional all core clock speed. Couple that with better memory access latency numbers due to a new IOD and inter die connection and I don't see where the need for N2 is really pressing.

The math ain't mathing for me...

The slide was just talking about IPC.

 

[adroc_thurston]

Going from N4P to N3P and including fin flex where it can help should be able to net a good 5+% clock speed

Whole lot more than that.
Going to N2p nets them like 20% speed with nano-flex usage.

 

[Kepler_L2]

Whole lot more than that.
Going to N2p nets them like 20% speed with nano-flex usage.

That would be 7.2 GHz from their original 6.0 GHz target for Zen4/5

 

[adroc_thurston]

That would be 7.2 GHz from their original 6.0 GHz target for Zen4/5

Remember that Vmax goes down on N3 and N2.

 

[Joe NYC]

Better, they like it in N2X flavour.

I hope Zen 6 is not late, to be able to use N2X

It is. For Medusa premium/halo.

So desktop IOD still on n6?

I wonder what percentage of desktop IOD (high) power consumption is SerDes and how much is just not enough power optimization on the IOD.

Server was the priority, and desktop was just a derivative up to now. But hopefully, with Zen 6, desktop IOD gets some power optimizations.

 

[adroc_thurston]

So desktop IOD still on n6?

probably N4c. Probably.

I wonder what percentage of desktop IOD (high) power consumption is SerDes and how much is just not enough power optimization on the IOD.

it's all the latter.

But hopefully, with Zen 6, desktop IOD gets some power optimizations.

gets all of them, it's also Gator Range.

 

[Tigerick]

That would be 7.2 GHz from their original 6.0 GHz target for Zen4/5

And remember that the core count has increased from 16 to 24, thus the benefits of new node most likely going to power additional 8 cores and L3 cache...Thus I am expecting big boost in MT scores but not ST. ST will depend on the final clock speed and IPC of the Zen6.

 

[inquiss]

Remember that Vmax goes down on N3 and N2.

So that's all base clock increase? Or, well, beneficial in laptop but not in desktop?

 

[adroc_thurston]

So that's all base clock increase?

Not really, no. They just get to ride the very end of the v/f curve a bit less.

Or, well, beneficial in laptop but not in desktop?

Yes, Medusa will be an xboxhueg upgrade for halo/luggable/premium t&l.

 

[basix]

It looks around 12% smaller. Using TSMC numbers for logic and SRAM scaling, and assuming 50/50 between Logic and SRAM, it's around 32% more transistors.

32% more transistors per core? That would be massive. Zen 4 & 5 got already much heavier compared to their predecessors.

That would be 7.2 GHz from their original 6.0 GHz target for Zen4/5

I would not mind getting Zen 6 with >7 GHz

I hoped that Zen 6 gets a bit past 6 GHz and I can operate it with 200-300MHz less & UV at at leat 6 GHz. Zen 6 + 6 GHz with very high efficiency just sounded nice. But if it's 7 GHz instead of 6 GHz? Well, shut up and take my money!

 

[yuri69]

32% more transistors per core? That would be massive. Zen 4 & 5 got already much heavier compared to their predecessors.

Not this hype again... The transition from 7nm to 5nm also brought "massive" amount of transistors for Zen 4 and we know how it went.

 

[basix]

More transistors != Hype, my friend

Transistor count goes up quite some bit and I do not correlate that to performance. I just wanted to point out that 32% is a very significant increase. Compared to Zen 3 we have have what, 3x more transistors per Core?

 

[Timorous]

Not this hype again... The transition from 7nm to 5nm also brought "massive" amount of transistors for Zen 4 and we know how it went.

Yea, massive clock speed gains and decent IPC gains that were worth a good 20% to 30% extra performance.

 

[adroc_thurston]

The transition from 7nm to 5nm also brought "massive" amount of transistors for Zen 4 and we know how it went.

29% 1t? Yeah that was pretty damn good. Next.

 

[StefanR5R]

There is a noticeable latency penalty caused by the SERDES, and it is basically the difference in memory latency between Intel and AMD. A wider interface can remove the need for serialization, and thus reduce the latency.

Most of the total access latency is DRAM latency, but that doesn't mean that the smaller parts don't matter. Latency is additive.

Apropos, are any figures out there anywhere about memory access latency of Ryzen 8000[F/G] compared to 7000/9000?

 

[LightningZ71]

Your math ain’t mathing for me, either.

Thanks for catching that...

 

[Hulk]

And remember that the core count has increased from 16 to 24, thus the benefits of new node most likely going to power additional 8 cores and L3 cache...Thus I am expecting big boost in MT scores but not ST. ST will depend on the final clock speed and IPC of the Zen6.

It is definite that core count is going from 16 to 24? I thought that was just something we were hoping for and seems likely?

 

[gdansk]

It is definite that core count is going from 16 to 24? I thought that was just something we were hoping for and seems likely?

There are some very realistic looking diagrams of 12 core (homogenous) CCD. It isn't confirmed by AMD but it seems likely.

 

[Joe NYC]

Not this hype again... The transition from 7nm to 5nm also brought "massive" amount of transistors for Zen 4 and we know how it went.

It went well for AVX512, it went well for SMT in server in MT, a little mixed for ST on client.

Which seems like a key thing for AMD to address in Zen 6.

Another thing (using up transistors) could be equivalent of Intel's AMX / SME instructions.

 

[Joe NYC]

A hypothesis based on how AMD is minting money on premium V-Cache processors, huge expectations for Strix Halo.

These two (plus Strix Point, another premium chip, doing less well) are all premium processors, and according to MLID leaks, all 3 of these premium product lines are going to be using the 12 core CCD, possibly N2.

Mainstream and low end CPUs will probably use different, likely monolithic dies.

Considering that all of these are premium products (not mass volume, higher price), then why not lock up the category by pairing all of them with V-Cache? What would be a better way to put Intel into Check - Mate then to knock out Intel completely from premium category of products?

If all of these N2 CCDs were to be paired with V-Cache, then removing 48 MB of L3 from main die to V-Cache wold be a non-brainer. From cost perspective:

- cost of 48 MB L3 on N2
vs.
-cost of adding N7 V-Cache die

could be a wash - while doubling the L3 to ~96 MB

There was another rumor floating around: that AMD would increase L2 from 1 MB to 2 MB. In case of 12 core CCD, that alone would be 24 MB of cache. Probably enough for lower end CPUs. But then, the lower end CPUs don't need 12 cores.

Only 1 segment would be missing a successor under this scenario - the low cost desktop 6 core CPUs, which could be addressed in other ways. And these chips, in any case, contribute the least to AMD gross margins.

 

[DavidC1]

29% 1t? Yeah that was pretty damn good. Next.

Yea, and the TDP went up from 105W to 170W despite moving to a new process. Pretty sure that has nothing to do with significant clock increase. If they can get significantly above 6GHz(which I have doubts), then they can have their own Raptorlake.

 

[gdansk]

Yea, and the TDP went up from 105W to 170W despite moving to a new process. Pretty sure that has nothing to do with significant clock increase. If they can get significantly above 6GHz(which I have doubts), then they can have their own Raptorlake.

How much did it go up in 1T? Not that much as I recall

Spoiler

AMD Ryzen 9 7950X Review - Impressive 16-core Powerhouse

The Ryzen 9 7950X is a monster CPU. When paired with the right workload it will eat even the 12900K for breakfast. As our review shows, the performance uplifts can be massive: +30-50% gen-over-gen is totally possible. What makes things complicated though, is that keeping the beast cool is almost...

www.techpowerup.com

They boosted the TDP because of their competition.

 

[DavidC1]

How much did it go up in 1T? Not that much as I recall

I just don't see why they want to keep focusing on clocks when it's clear ARM competition needs nowhere high for equal, or even better performance.

The M4 max with 14P cores is just 20% away from the monsters like 285K and 9950X. And it that in a slim laptop form factor using 57W. And in single thread, at 4.6GHz it's beating 5.7GHz chips by 20%.

Despite the financial highs they are experiencing right now, it tells me AMD is also going to face a point where the deficit will catch up to them and will start a decline. If people think AMD's chip design is at peak execution, they are sorely mistaken.

 

[gdansk]

I just don't see why they want to keep focusing on clocks when it's clear ARM competition needs nowhere high for equal, or even better performance.

The M4 max with 14P cores is just 20% away from the monsters like 285K and 9950X. And it that in a slim laptop form factor using 57W. And in single thread, at 4.6GHz it's beating 5.7GHz chips by 20%.

Despite the financial highs they are experiencing right now, it tells me AMD is also going to face a point where the deficit will catch up to them and will start a decline. If people think AMD's chip design is at peak execution, they are sorely mistaken.

Not the point. Zen 4 wasn't bad, clock speed worked for it.

And regarding Zen 6 we have good evidence suggesting IPC increase is generally meagre. So what's left to hope for? 6.2 GHz peak 1T and more cores.

 

[adroc_thurston]

Yea, and the TDP went up from 105W to 170W despite moving to a new process

Eh it was struggling to actually hit the rated 230W PPT.

I just don't see why they want to keep focusing on clocks when it's clear ARM competition needs nowhere high for equal

Because you're literally like a year away from aa64 camp putting out 5Ghz chips.

If people think AMD's chip design is at peak execution, they are sorely mistaken.

Indeed, Venice is helluva lot cooler than Turin.

And regarding Zen 6 we have good evidence suggesting IPC increase is generally meagre

You'll get 15% and you're gonna like it.