Question Zen 6 Speculation Thread

Page 196 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Makaveli

Diamond Member
Feb 8, 2002
4,966
1,561
136
OEMs are probably only buying the "low end" models.
if you only focus on laptops and apu's maybe.

They sell more than that example Lenovo I bought 4 TR workstations from them a few years ago for work.

When you view things from the corporate side and not just us building our own machines, we don't get custom orders and special silicon just for us they do.

however no follow up post on this from me I don't want to go off topic.
 
Last edited:
  • Like
Reactions: Io Magnesso

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
27,097
16,014
136
What I was referring to in the 40% I was talking about was Zen 4 to Zen 5, and I guess that was actually based on the server CPUs by adroc. I have found the same in 9950x vs 7950x in very heavy avx-512 apps.
 

StefanR5R

Elite Member
Dec 10, 2016
6,555
10,305
136
The 40% thing was related to SPECrate® 2017 Integer, it was guesswork based on preliminary info on per-socket performance of top-end server, and as such this guesswork turned out wrong. Any similarities of these obsolete speculations to unrelated figures in the 40% ballpark are mere coincidence.
 

Timorous

Golden Member
Oct 27, 2008
1,966
3,850
136
What I was referring to in the 40% I was talking about was Zen 4 to Zen 5, and I guess that was actually based on the server CPUs by adroc. I have found the same in 9950x vs 7950x in very heavy avx-512 apps.

There was this slide.

I have found the 32%. Not Specint though.

mark-papermaster-11.jpg
 

Bigos

Member
Jun 2, 2019
198
514
136
There is no news to be had there, while are you still linking his videos?

And the question? Are we going to watch the video for you? I thought you were supposed to put commentary with a video.

This looks like you are actively promoting MLID to have more views.
 
Jul 27, 2020
26,084
17,983
146
Dan, how are you doing?
I'm pretty good. How about you?
Yeah, I'm doing pretty well. Just before I came upstairs, I was downstairs getting some tea when I noticed my girlfriend fiddling with the remote. She said, "I don't know why the TV was on when I came in," and then I saw Jesse on the ground next to the remote. Now, I can't get the remote to work. I thought, "Great, guess I'll go to work." But she just texted me saying it was the batteries — she probably knocked them loose when she jumped on the coffee table, which she does to look out the windows, as you've seen. I was watching her take out the trash or do yard work because she has to know what we're doing at all times. Our dog. She probably knocked the remote off the coffee table, which caused a bit of stress. Also, I noticed the TV channel suspiciously turned to Melanino videos or something similar — just cops or firefighters doing their thing — or maybe a rerun of "Cops." She mentioned, "Belgian Malinois are like combat dogs that all police and enforcement agencies use now," which is hilarious because it's clearly in her blood. If a firefighter, cop, or confident mail carrier passes by, she's on board and watches them like a spy. We’re in Nashville, so we often see musicians walking by — she doesn’t seem to like them. Sometimes they park near us to go to a gig or restaurant, and she barks at them. Recently, there was an incident nearby with cops checking out something, and she was fine with the police walking around, even smiling and asking if she could help.


Anyway, since this is a gaming hardware podcast, we should probably move on from dog talk. But, based on the comments, community feedback indicates that dog and cat talk is generally welcomed, with only a few disliking it. I remember early on, some people complained about dogs on screen during a live stream — I think back when I lived in Peoria. Someone said, “Dogs are disgusting,” in the chat. I found that interesting, but we’re here for the hardware content, so no worries. Just keep in mind, most of the community prefers animals being part of the vibe. The point is, after six years, we're still waiting for a truly compelling GPU upgrade in the 60-series; the performance uplift has stagnated. Nvidia's 3060, 4060, and 5060 series have shown slow or no significant gains. The 8GB models are outsold by 16GB versions at around a 16:1 ratio, which is roughly a 1600% difference — indicating gamers are rejecting the 8GB variants. Recent reviews from Hardware Unboxed confirm this: Nvidia launched the RTX 5060 Ti with driver restrictions, so initial benchmarks were limited, but even then, performance gains over the 2060 were modest — around 20-25%. The 3060, with 12GB, still outperformed the 8GB 4060 in many scenarios, highlighting that VRAM alone isn't enough. The 5060 series, despite being priced similarly to the 4060, offers about 20% better value but still underwhelms. After six years since the 2060, we’re still waiting for a GPU that offers a strong performance boost at a fair price with ample VRAM — ideally 12-16GB. Nvidia’s slow progress indicates a decline in what these models represent; they should be called "50" series now. Meanwhile, the market response is clear: gamers are voting with their wallets, and the trend favors higher VRAM options, especially as the 60-class has become less relevant. This stagnation might push Nvidia to respond with a true next-gen product.


In other news, Nvidia's upcoming RTX 5090 or 9090 series is expected to surpass 92 TFLOPS in FP32 performance, potentially exceeding 7 GHz clock speeds, based on leaks from multiple sources. AMD’s Zen 6 architecture is also confirmed by leakers: it’s an evolution of Zen 5, with up to 12 cores per CCD, totaling 24 cores and 48 threads on a desktop. It’s expected to maintain compatibility with existing AM5 motherboards, with no major changes to boost algorithms or curve optimizer. Internally, AMD is testing Zen 6 CPUs reaching 6.4 GHz, aiming to surpass 7 GHz on the high-end TSMC N2X process — an unprecedented feat in x86 history. Mobile versions might reach around 6.2-6.5 GHz. The die size is estimated at around 75mm², roughly a quarter of Zen 4 or Stricks’ size, with a small integrated GPU (likely 1-2 RDNA 4 compute units). Pricing per core is expected to decrease, but AMD will probably charge more for higher tiers, maintaining or slightly raising prices. For example, a 10-core chip might be around $550, offering a significant performance uplift. AMD’s strategy appears to be a gradual progression rather than a revolutionary overhaul.


Regarding AMD’s future, with Zen 7, they plan to add support for AVX-10 and possibly other advanced extensions, pushing further into high-frequency, high-performance territory. Meanwhile, Intel’s future looks bleak: their CEO, Pat Gelsinger, reportedly considers the company’s hardware ambitions dead, acknowledging they are no longer a top 10 chipmaker, and that their struggles are due to internal shortcomings and failure to keep pace, especially in AI. Multiple studio layoffs have occurred, and there are rumors that Phil Spencer may retire after the next Xbox console launches, with Sarah Bond poised to take over. Microsoft’s Xbox division appears to be in decline, with former insiders claiming the hardware and backward compatibility efforts are waning, and the company’s focus shifting away from dedicated gaming hardware to PC-centric services. Phil Spencer’s departure seems imminent, but Microsoft may try to spin it as a retirement for image reasons. The overall outlook for Xbox hardware and dedicated gaming consoles is grim, and Microsoft’s strategy seems uncertain.


In the graphics market, Nvidia’s new 50-series GPUs, like the RTX 5060 Ti, are underperforming and overpricing. The 8GB models are outsold by 16GB variants at a 16:1 ratio, reflecting gamer rejection. The performance gains over prior generations are minimal, around 20% every two years, which is disappointing. AMD’s RDNA 3.5 architecture (e.g., the 7600 series) is better than RDNA 3 but still not at the level of RDNA 4, which offers significant improvements, especially in ray tracing. AMD needs to accelerate development of low-end RDNA 4 cards with at least 12-16GB VRAM to stay competitive. In mobile, Nvidia’s RTX 5090 series is likely to skip a refresh if AMD doesn’t introduce new high-performance parts soon, as Nvidia currently dominates laptop graphics.


On the platform and infrastructure front, AMD is collaborating closely with Sony on PS6 and PS5 Pro, with new advanced upscaling tech (FSR4) being developed to enhance graphics performance. Mark Cerny, lead architect at Sony, stated that AMD and Sony are working together on integrating these technologies, which will be available to all AMD’s customers. The new RDNA 5 architecture is heavily influenced by this joint effort, possibly incorporating features like more demanding scene rendering and higher clock speeds. There are also indications that the next-generation consoles may leverage this collaboration for better performance and compatibility, with AMD’s RDNA 5 expected to include substantial contributions from Sony’s engineering teams.


Looking ahead, the industry faces a slow decline in PC gaming hardware innovation, with Nvidia’s 60-series GPUs stagnating in performance and VRAM, and AMD still catching up at the low end. Meanwhile, Sony is pushing its next-gen console tech and working closely with AMD, hinting at a future where console and PC graphics are increasingly aligned.


Finally, other updates include:


  • TSMC reportedly has hit maximum capacity for 2nm manufacturing, which may delay some upcoming AMD and Intel products.
  • Leaked specs of Intel’s Diamond Rapids CPU show it could reach 192 cores, confirming long-standing rumors.
  • LPDDR6 memory is officially announced and will be used in upcoming AMD APUs like Medusa Halo, possibly arriving around 2027, which explains the long delays for Halo’s release.
  • The upcoming Nintendo Switch 2 is experiencing ongoing shortages, with Nintendo often limiting supply deliberately — likely to boost demand for a future model, possibly a Switch OLED revision.
  • The gaming market continues to shift towards online-only models, with concerns about game preservation and ownership. Legislation in the EU might require publishers to ensure long-term access to purchased games, reducing the risk of digital content becoming inaccessible due to store closures.

Overall, the industry faces a period of stagnation and strategic realignment, with AMD making notable advancements, Nvidia’s offerings stagnating, and Microsoft’s gaming division possibly in decline. Support for independent, leak-based journalism remains vital to stay informed of these ongoing developments.
 

511

Platinum Member
Jul 12, 2024
2,961
2,951
106
Nvidia's upcoming RTX 5090 or 9090 series is expected to surpass 92 TFLOPS in FP32 performance, potentially exceeding 7 GHz clock speeds, based on leaks from multiple sources

Meanwhile, Intel’s future looks bleak: their CEO, Pat Gelsinger, reportedly considers the company’s hardware ambitions dead, acknowledging they are no longer a top 10 chipmaker, and that their struggles are due to internal shortcomings and failure to keep pace, especially in AI.

The die size is estimated at around 75mm², roughly a quarter of Zen 4 or Stricks’ size, with a small integrated GPU (likely 1-2 RDNA 4 compute units). Pricing per core is expected to decrease, but AMD will probably charge more for higher tiers, maintaining or slightly raising prices.
Some AI generated Highlights

MLID+AI = Record Breaking Leaks
 

Josh128

Golden Member
Oct 14, 2022
1,060
1,608
106
There is no news to be had there, while are you still linking his videos?

And the question? Are we going to watch the video for you? I thought you were supposed to put commentary with a video.

This looks like you are actively promoting MLID to have more views.
Still linking? Where else have I linked his videos? Its better than the previous page of OT AI crap that has nothing to do with Zen 6. At least its supposed to contain info about Zen 6.

Thats the idea, I didnt have time to watch it. My post does contain a comment, dont you see it?

If thats what it looks like to you, you need to have your eyes checked.

**EDIT: Now that Ive actually watched the part about Zen 6, it seems hes now claiming its been confirmed they've already achieved 6.4GHz and that the number should only go up in the final product. I have no idea where hes pulling that "confirmation" from, but it does look to me like an attempt to hedge his bets and not look like a total fraud when the chips come out and max out there or a bit less.
 
Last edited:
Jul 27, 2020
26,084
17,983
146
**EDIT: Now that Ive actually watched the part about Zen 6, it seems hes now claiming its been confirmed they've already achieved 6.4GHz and that the number should only go up in the final product. I hae no idea where hes pulling that "confirmation" from, but it does look to me like an attempt to hedge his bets and not look like a total fraud when the chips come out and max out there or a bit less.
I'm guessing 6.3 GHz minimum to take the crown from 14900KS.

Would be crazy if 10600X hits those speeds too.
 
  • Like
Reactions: Tlh97 and Joe NYC

511

Platinum Member
Jul 12, 2024
2,961
2,951
106
Still linking? Where else have I linked his videos? Its better than the previous page of OT AI crap that has nothing to do with Zen 6. At least its supposed to contain info about Zen 6.

Thats the idea, I didnt have time to watch it. My post does contain a comment, dont you see it?

If thats what it looks like to you, you need to have your eyes checked.

**EDIT: Now that Ive actually watched the part about Zen 6, it seems hes now claiming its been confirmed they've already achieved 6.4GHz and that the number should only go up in the final product. I have no idea where hes pulling that "confirmation" from, but it does look to me like an attempt to hedge his bets and not look like a total fraud when the chips come out and max out there or a bit less.
whenever he says something like this remember 40% improvement for ARL vs MTL
 

LightningZ71

Platinum Member
Mar 10, 2017
2,322
2,913
136
Either they are using the absolute loosest bins possible to get the maximum number of viable dice per wafer, or they used the densest possible libraries to maximize the wafer area usage by having smaller die and more of them but sacrificing clock speed in the process. Not a problem in this segment really.
 
  • Like
Reactions: Io Magnesso

Gideon

Platinum Member
Nov 27, 2007
2,013
4,992
136
What evidence do we have that they ever double stacked for EPYC? Someone mentioned this earlier in the thread as if it was common knowledge.
There 100% was an early EPYC BIOS image that had options for enabling 2x and 4x stacks.

I don't remember if it was Servethehome, twitter gang or this very forum i saw it in. My google-fu has failed me repeatedly in finding it but it was a quite reputable source.
 

soresu

Diamond Member
Dec 19, 2014
3,898
3,331
136
LLMs are a force multiplier for R&D and all knowledge work pushing boundaries

expect the zen6 generation jump to have significant improvements over most other jumps

but also other companies will with N2 will catch up
AI/ML is a consideration for CPU µArch design, but compared to GPU µArch it's relatively minor.