Question Zen 6 Speculation Thread

[OneEng2]

You do realize Zen 5 was a complete flop? It took the x3D chips to save it. On the server side, excellent. On the PC side, awful release. AMD should use N3P for Zen 6 and make the x3D chips standard. AMD should consider releasing a Zen 5+ on 3nm mid 2025. N3P for Zen 6 and Zen 7 would be the play for AMD.

In what way do you consider Zen 5 a flop? AMD is absolutely going to use N3P for Zen 6 from what I hear. Not sure about x3D, but considering that it only helps in part of the market, I am not sure I agree.

There is no reason for AMD to release another Zen 5 variant IMO. Better to wait on N3P and N2 for higher transistor density and actually add some IPC and cores to the mix.

Intel wishes it had a complete flop that profitable.

AMD has continued to gain market share in desktop and server with Zen 5. Seems like a pretty successful product to me.... and they are doing it on a less expensive, less dense node that Intel's Arrow Lake.

Last time I saw Panther Lake is still behind Strix Point especially in MT, maybe close to double digit at last. ST might be a tie.

Up to 18 cores (6P 8E and 4LPE). Seems like it would be pretty potent to me. Of course, without SMT, you are looking at 18 cores and 18 threads vs 16 cores and 32 threads (halo) and 12 cores and 24 threads for Strix Point. Considering the use of BSPDN on 18A, the actual transistor count could be significantly increased over Intel's current generation giving them the opportunity for some serious IPC uplift .... so I am not sure I agree with your assessment.

 

[Kepler_L2]

Up to 18 cores (6P 8E and 4LPE)

It's 4+8+4

 

[OneEng2]

It's 4+8+4

Seems like there is talk of a 6+8+4 version as well.

 

[vanplayer]

Up to 18 cores (6P 8E and 4LPE). Seems like it would be pretty potent to me. Of course, without SMT, you are looking at 18 cores and 18 threads vs 16 cores and 32 threads (halo) and 12 cores and 24 threads for Strix Point. Considering the use of BSPDN on 18A, the actual transistor count could be significantly increased over Intel's current generation giving them the opportunity for some serious IPC uplift .... so I am not sure I agree with your assessment.

There were 2 narratives: 4+8 or 6+8. The performance estimated on Pantherlake is for 6+8 part, but I'm not sure if the source misunderstand the pantherlake spec.
4+8(+4) being double digit slower than Strixpoint sounds right, if it's 6+8(+4) being slower that much I think something is horribly wrong.

sry I forget this is AMD thread.

 

[yuri69]

There is no reason for AMD to release another Zen 5 variant IMO. Better to wait on N3P and N2 for higher transistor density and actually add some IPC and cores to the mix.

Strix Point has been released in June 2024 and its Zen 6-based replacement comes in June 2026 or maybe even Jan 2027. That's quite a gap for a "premium" product.

The 2 year release cycle forced AMD to release Zen 3+-based Rembrandt, so I would expect something better than just a plain Strix Point rebrand in 2025.

 

[OneEng2]

There were 2 narratives: 4+8 or 6+8. The performance estimated on Pantherlake is for 6+8 part, but I'm not sure if the source misunderstand the pantherlake spec.
4+8(+4) being double digit slower than Strixpoint sounds right, if it's 6+8(+4) being slower that much I think something is horribly wrong.

sry I forget this is AMD thread.

Ahh. Yes, I agree. If there were 18 full cores, I might expect to best strix point easily in MT.

Strix Point has been released in June 2024 and its Zen 6-based replacement comes in June 2026 or maybe even Jan 2027. That's quite a gap for a "premium" product.

The 2 year release cycle forced AMD to release Zen 3+-based Rembrandt, so I would expect something better than just a plain Strix Point rebrand in 2025.

Sadly, I think we should all get used to a much slower release cadence in the future. It doesn't make sense to do a major redesign without a higher transistors budget. Sure, we can expect a tweak or two between redesigns, but it isn't like there is this vast pool of performance tweaks that were simply overlooked in the initial design. Let's give the engineers some credit eh?

 

[Win2012R2]

It doesn't make sense to do a major redesign without a higher transistors budget

Costs more than consumer market prepared to pay - even without "AI" competition for top end silicon process. Where are those fancy graphene transistors clocking 50 Ghz when you need 'em...

 

[yuri69]

Sadly, I think we should all get used to a much slower release cadence in the future. It doesn't make sense to do a major redesign without a higher transistors budget. Sure, we can expect a tweak or two between redesigns, but it isn't like there is this vast pool of performance tweaks that were simply overlooked in the initial design. Let's give the engineers some credit eh?

Hmm, OEMs are the ones who decide things here - they need to put something fresh-looking on the shelves every year. That's why we see those stupid rebrands of rebrands.

AMD introduced a "+ model" for both previous major architectural overhauls of the Zen core IP. Zen 1+ was a classic bugfix release of Zen 1. Zen 3+ was a power-mngmnt fix release of Zen 3. I quite hope they are gonna perform a similar patching session for Zen 5 in 2025 instead of a dull rebrand.

 

[adroc_thurston]

I quite hope they are gonna perform a similar patching session for Zen 5 in 2025 instead of a dull rebrand.

Zen5+ is years like dead.

OEMs are the ones who decide things here - they need to put something fresh-looking on the shelves every year

Even Intel is struggling to run yearly in client. They gotta chill down.

 

[Tuna-Fish]

Where are those fancy graphene transistors clocking 50 Ghz when you need 'em...

Fyi, the silicon transistors in a cpu need to be able to switch at over 100GHz for the cpu to reach 6GHz. Cpus don't just have individual transistors switching as fast as they can, they have trees of transistors driving each other, and if a tree is 20 transistors long, individual transistors in a tree need to be able switch 20 times faster than the whole thing.

50GHz really isn't very fast for single-transistor performance these days.

 

[OneEng2]

Hmm, OEMs are the ones who decide things here - they need to put something fresh-looking on the shelves every year. That's why we see those stupid rebrands of rebrands.

AMD introduced a "+ model" for both previous major architectural overhauls of the Zen core IP. Zen 1+ was a classic bugfix release of Zen 1. Zen 3+ was a power-mngmnt fix release of Zen 3. I quite hope they are gonna perform a similar patching session for Zen 5 in 2025 instead of a dull rebrand.

I think that is much more likely than a new core. Just a tweak here and there where things really sucked.

 

[Joe NYC]

Strix Point has been released in June 2024 and its Zen 6-based replacement comes in June 2026 or maybe even Jan 2027. That's quite a gap for a "premium" product.

The 2 year release cycle forced AMD to release Zen 3+-based Rembrandt, so I would expect something better than just a plain Strix Point rebrand in 2025.

Being between Kraken and Strix Halo, Strix Point diminishes in its importance. Many of the Strix Point models are paired with mobile NVidia 4060 and 4050, which should be mostly eclipsed by Strix Halo - if the there is market reception for it and AMD plays its cards right.

There were some rumors of a possible Strix Halo successor before Medusa Halo. Which is something that would be needed much more urgently, given NVidia is going to enter the market with Strix Halo competitor.

 

[Meteor Late]

Laptops with RTX 4060 can be regularly seen for 1000€ or lower here in Europe or in U.S.
We will see who will exactly buy Strix Halo and what price it will have, because as a gaming device it doesn't really seem enticing to me if the rumors that it will be expensive, like 1600$+ expensive, are true.

 

[marees]

Laptops with RTX 4060 can be regularly seen for 1000€ or lower here in Europe or in U.S.
We will see who will exactly buy Strix Halo and what price it will have, because as a gaming device it doesn't really seem enticing to me if the rumors that it will be expensive, like 1600$+ expensive, are true.

I don't think laptop customers decide GPU.

More like Asus who is creating the RoG Flow for example

 

[CouncilorIrissa]

Even Intel is struggling to run yearly in client. They gotta chill down.

Arm/Qualcomm isn't. At this rate 2026 is going to look real awkward for AMD if their supposed premium laptops are losing to flagship phones in 1t by 15-20%. And don't get me started on a potential SDXE gen 2.

 

[StefanR5R]

if their supposed premium laptops are losing to flagship phones in 1t by 15-20%

Before this happens, premium laptop makers could hire firmware programmers off of flagship phone makers, so that premium laptops too become able to switch into meltdown mode when the run of a popular benchmark program is detected.

 

[adroc_thurston]

Arm/Qualcomm isn't

Yea they are, QC laptop parts aren't 4Q cadence either.

 

[fastandfurious6]

merry christmas zen speculators!

 

[OneEng2]

Merry Christmas!

 

[DavidC1]

Considering the use of BSPDN on 18A, the actual transistor count could be significantly increased over Intel's current generation giving them the opportunity for some serious IPC uplift .... so I am not sure I agree with your assessment.

Pantherlake is a Tick with supposedly 8% improvement on the P core and 3-5% on the E core. Only clocks will make the difference, and with that much of a thread count, it won't be enough.

 

[OneEng2]

Pantherlake is a Tick with supposedly 8% improvement on the P core and 3-5% on the E core. Only clocks will make the difference, and with that much of a thread count, it won't be enough.

Perhaps the best benefit will be to Intel's costs then?

 

[Win2012R2]

Perhaps the best benefit will be to Intel's costs then?

It might be interesting if they get more than 8 P=cores on one chiplet - they need it for Zen 6 competition.

 

[OneEng2]

It might be interesting if they get more than 8 P=cores on one chiplet - they need it for Zen 6 competition.

The rumors I have heard have desktop Zen 6 using 12 core CCD's (vs 8 today). Considering that I haven't seen a single shred of information that suggests that Cougar Cove will have SMT, I am thinking that they will need quite a few cores to compete with Zen 6.

It is interesting to me that Intel is stratifying their cores further. Now, instead of just P and E, we have E low power as well. It will be interesting to see how this pans out.

 

[inquiss]

The rumors I have heard have desktop Zen 6 using 12 core CCD's (vs 8 today). Considering that I haven't seen a single shred of information that suggests that Cougar Cove will have SMT, I am thinking that they will need quite a few cores to compete with Zen 6.

It is interesting to me that Intel is stratifying their cores further. Now, instead of just P and E, we have E low power as well. It will be interesting to see how this pans out.

So is AMD...

 

[Thibsie]

The rumors I have heard have desktop Zen 6 using 12 core CCD's (vs 8 today). Considering that I haven't seen a single shred of information that suggests that Cougar Cove will have SMT, I am thinking that they will need quite a few cores to compete with Zen 6.

It is interesting to me that Intel is stratifying their cores further. Now, instead of just P and E, we have E low power as well. It will be interesting to see how this pans out.

Yeah but... 12 Zen6 cores or 6 Zen6 + 6 Zen6c cores or variations of those ?