Question Zen 6 Speculation Thread

[MoistOintment]

It makes sense why AMD is moving to 12 core CCDs in server, and thus desktop also becomes 12 core CCDs as a byproduct. They didn't need a core count increase in client.

But now that 11950X (or whatever it's called) will be 24 cores, 485K (or whatever it's called) would lose bad in high nT productivity. The only response Intel could possibly have is either accept nT productivity defeat by a wide margin, or add a second compute tile and bring core count up to 48 cores.

Yes, these very high core count CPUs today are a bit gimmicky and there's a small niche that would truly benefit. And Yes, given the choice, I think the vast majority would prefer big ST improvements. But if ST improvements are slowing, then it makes sense to push core counts.

The point is that the halo effect of having "the best" CPU has a big impact on down stream sales. Even amongst the "enthusiast" community. And those enthusiasts act as "tech advisors" to non-techy friends and family.

There's a big downstream benefit to having the "best" client CPU at gaming and productivity, even it itself doesn't contribute a lot financially.

 

[gdansk]

That is harmful generalization. It all depends on what you are doing.

Yeah, but we're talking about chips that are going to have a measly meager paltry 24 cores. The amount of people who need more but don't want more memory bandwidth is small. Perhaps almost the entire market is reading this thread conjuring imaginary workloads where 12 Z6 CCD + 32 Z6c CCD would be a good configuration.

 

[Hulk]

It makes sense why AMD is moving to 12 core CCDs in server, and thus desktop also becomes 12 core CCDs as a byproduct. They didn't need a core count increase in client.

But now that 11950X (or whatever it's called) will be 24 cores, 485K (or whatever it's called) would lose bad in high nT productivity. The only response Intel could possibly have is either accept nT productivity defeat by a wide margin, or add a second compute tile and bring core count up to 48 cores.

Yes, these very high core count CPUs today are a bit gimmicky and there's a small niche that would truly benefit. And Yes, given the choice, I think the vast majority would prefer big ST improvements. But if ST improvements are slowing, then it makes sense to push core counts.

The point is that the halo effect of having "the best" CPU has a big impact on down stream sales. Even amongst the "enthusiast" community. And those enthusiasts act as "tech advisors" to non-techy friends and family.

There's a big downstream benefit to having the "best" client CPU at gaming and productivity, even it itself doesn't contribute a lot financially.

The problem the "winner" of the MT wars is going to have is finding benchmarks/applications that will actually sell people on the need for such high core counts.
Beyond 24/48 I don't think many apps are going to show improvements.

Games? No.
Photoshop? No.
Video encoding? Not really, perhaps just a bit with some formats.
Business? No, sorry.
Web Browsing? Haha.
Cinebench and similar? Yes. But this isn't what "the people" are generally interested in for actual work.

So, assuming plenty of cores for each, it's gonna be the IPC and clocks those first handful of cores can reach to determine the better CPU.

 

[511]

We need a 10 GHz Zen 6 .

 

[Josh128]

Yeah, but we're talking about chips that are going to have a measly meager paltry 24 cores. The amount of people who need more but don't want more memory bandwidth is small. Perhaps almost the entire market is reading this thread conjuring imaginary workloads where 12 Z6 CCD + 32 Z6c CCD would be a good configuration.

Dont forget, add X3D to both of those CCDs. Would be even better.

 

[gdansk]

Dont forget, add X3D to both. Would be even better.

I heard one of the area savings the c cores had made also mean it is impossible to attach vertical cache. But who knows.

 

[Hulk]

We need a 10 GHz Zen 6 .

Intel... "Been there. Done that. 20 years ago."

On paper anyway.

 

[Hulk]

This is what chatgpt (which if of course never wrong) says about frequency if the transistors could operate at light speed.

Scale / Distance	Frequency (Hz)	In GHz	Order-of-magnitude

5 nm gate

6 × 10¹⁶ Hz

6 × 10⁷ GHz

≈ 60 million GHz

100 nm interconnect

3 × 10¹⁵ Hz

3 × 10⁶ GHz

≈ 3 million GHz

1 µm block

3 × 10¹⁴ Hz

3 × 10⁵ GHz

≈ 300 thousand GHz

1 cm chip

3 × 10¹⁰ Hz

30 GHz

≈ 30 GHz

 

[Josh128]

I heard one of the area savings the c cores had made also mean it is impossible to attach vertical cache. But who knows.

Never heard this, but it sounds likely.

 

[511]

This is what chatgpt (which if of course never wrong) says about frequency if the transistors could operate at light speed.

Scale / Distance	Frequency (Hz)	In GHz	Order-of-magnitude

5 nm gate

6 × 10¹⁶ Hz

6 × 10⁷ GHz

≈ 60 million GHz

100 nm interconnect

3 × 10¹⁵ Hz

3 × 10⁶ GHz

≈ 3 million GHz

1 µm block

3 × 10¹⁴ Hz

3 × 10⁵ GHz

≈ 300 thousand GHz

1 cm chip

3 × 10¹⁰ Hz

30 GHz

≈ 30 GHz

ChatGPT yeah totally wrong

 

[gdansk]

But now that 11950X (or whatever it's called)

I suspect they'll call it 595X. A 10950X or 11950X might almost make sense, so AMD will avoid that.

 

[Josh128]

Okay prediction time.

Based on my rig running CB R23 (more fun to run/watch than R24) as the baseline for Zen 6 predictions....
This is how I've been running (and I hit it hard) for the past year with absolutely no issues and perfect stability.
For the record, I think it's nearly impossible for Raptor Lake to attain this score without a custom loop and be reliable.
That's why after 30+ years of Intel I moved to AMD.
Anyway...

9950X
4.87GHz CCD1, 4.55 CCD2 - Average clockspeed = 4.71GHz
200W Package Power
75 C on 280 AIO
1.09V
CB R23 MT score 41,000

What will the clockspeed and CB R23 score be for the 24/48 Zen 6 CPU be at 200W?

If AMD does not set Zen 6 at 200W "out of the box" then what power will they use and what will the frequency and score for CBR23 be?

I'm curious for the 6 to 7GHz people out there how many will be claiming over 6GHz in this actual real world example for Zen 6? Even if you shut down 8 cores and made it apples-to-apples what do you think for frequency and score?

~45% power savings per core at same freq vs N4P. You have 50% more cores, so you would end up around 235MHz less nT clockspeed in this workload at the same 200W PPT. At the same time, lets say you gain 5% IPC in Cinebench R23. The results are:

For 24 core--
@200W :
41K*1.5 = 61,500 R23 nT
4.47 GHz average nT clockspeeds

@230W
67,000 R23 nT
4.9GHz average clockspeeds

4.47 GHz leaves a TON of room for more perf if you up the power, which is why I say they will-- they will pin that baby to whatever the socket can safely provide, at least 230W, maybe >250W.

 

[Josh128]

Hmm...does this also mean Venice launching before Medusa??

AMD Confirms openSIL Support For Zen 6 Ryzen "Medusa" CPUs In 1H 2027, EPYC "Venice" In 2026

AMD has confirmed its commitment to openSIL "Open Firmware" for next-gen Zen 6-based Ryzen "Medusa" & EPYC "Venice" CPUs.

wccftech.com

 

[StefanR5R]

I'm curious for the 6 to 7GHz people out there how many will be claiming over 6GHz in this actual real world example [CB23 MT] for Zen 6?

Originally, when the 6 GHz < f < 7 GHz discussion started,

• f was f_max, and
• "as close to 7 GHz as [super]humanly possible" was pictured as a development stretch goal given to AMD's [super]human engineers by their superiors.

Subsequently, various people here and at worse educated places throughout the 'net grossly mistook the f for average clocks under high CPU utilization. Furthermore, there were some who misrepresented the claims concerning those 7 GHz as claims that 7 GHz will be the f_max — and after having misrepresented the claims, could conveniently ridicule the claims.

PS, IIRC, CB23 is relatively light on memory accesses (lighter than CB24 which still isn't particularly heavy on mem access), therefore CB23 represents only a niche within the real world of parallel computing.

 

[Kaffeekenan]

Hmm...does this also mean Venice launching before Medusa??

AMD Confirms openSIL Support For Zen 6 Ryzen "Medusa" CPUs In 1H 2027, EPYC "Venice" In 2026

AMD has confirmed its commitment to openSIL "Open Firmware" for next-gen Zen 6-based Ryzen "Medusa" & EPYC "Venice" CPUs.

wccftech.com

Hm, MLID said "late 2025 or early 2026" for Venice. At least according to him the answer would be a resounding "yep".

 

[StefanR5R]

Hmm...does this also mean Venice launching before Medusa??

https://wccftech.com/amd-confirms-opensil-support-zen-6-ryzen-medusa-cpus-1h-2027-epyc-venice-2026/

No, launch order or launch dates cannot be derived from this information. Client CPU products can (and I presume will) launch with classic AGESA. Maybe openSIL never even gets to client AM5 platforms [edit: and laptop BGA sockets] and is merely adopted in server AM5 platforms [edit: or/and embedded platforms] besides SP7/SP8.

 

[Saylick]

Okay prediction time.

Based on my rig running CB R23 (more fun to run/watch than R24) as the baseline for Zen 6 predictions....
This is how I've been running (and I hit it hard) for the past year with absolutely no issues and perfect stability.
For the record, I think it's nearly impossible for Raptor Lake to attain this score without a custom loop and be reliable.
That's why after 30+ years of Intel I moved to AMD.
Anyway...

9950X
4.87GHz CCD1, 4.55 CCD2 - Average clockspeed = 4.71GHz
200W Package Power
75 C on 280 AIO
1.09V
CB R23 MT score 41,000

What will the clockspeed and CB R23 score be for the 24/48 Zen 6 CPU be at 200W?

If AMD does not set Zen 6 at 200W "out of the box" then what power will they use and what will the frequency and score for CBR23 be?

I'm curious for the 6 to 7GHz people out there how many will be claiming over 6GHz in this actual real world example for Zen 6? Even if you shut down 8 cores and made it apples-to-apples what do you think for frequency and score?

Sure I'll throw my hat into the ring.

200W, 16 core Zen 6, CB R23 MT, average clocks of ~5.5 GHz.

 

[Hulk]

Sure I'll throw my hat into the ring.

200W, 16 core Zen 6, CB R23 MT, average clocks of ~5.5 GHz.

Nice. How about a score?

 

[Saylick]

Nice. How about a score?

No idea.

 

[511]

AMD will call it Ryzen AI 9 590X

 

[gdansk]

Nobody is forcing you to buy 48c part if you can do everything you want with 8c machine.

Oh, I forgot to reply to this part too. We already saw how it played out with AMD. If you want the best single compute tile part, you can't buy it. AMD won't make it. It's not a big deal - maybe 3-4% slower than it could be.

But I suspect Intel will do the same. If dual tile NVL didn't exist you can guarantee that their fastest, best bin chips would appear in a single tile configuration. The mere existence of a dual tile configuration gives Intel an option to hurt every other buyer down the stack in the name of product segmentation to push that ASP.

 

[DrMrLordX]

Many seem to assume that ARL is a "dud" and brush aside NVL.

Intel keeps recycling the same basic P-core design with minor tweaks. Until they trot out unified core, nobody seems to expect much of them.

Beyond 24/48 I don't think many apps are going to show improvements.

You could argue that 16c/32t is already at that point.

 

[511]

High core count are going to be pricey anyway real pricey.

 

[511]

Intel keeps recycling the same basic P-core design with minor tweaks. Until they trot out unified core, nobody seems to expect much of them.

Lion Cove is not a minor tweak ... unlike GLC

 

[Josh128]

Oh, I forgot to reply to this part too. We already saw how it played out with AMD. If you want the best single compute tile part, you can't buy it. AMD won't make it. It's not a big deal - maybe 3-4% slower than it could be.

Well, if the 9850X3D rumor of 5.6GHz is true, you WILL be able to buy the best single compute tile part. Im just still skeptical its actually real.