Question Zen 6 Speculation Thread

[adroc_thurston]

when 3nm wafer supply is going to open up.

it ain't, MDS1/2/3 are there.

 

[Fjodor2001]

It's a drop-in upgrade for the existing AM5 install base.

Hvabehar? What does that have to do with release dates on DT etc.

 

[Fjodor2001]

I explained exactly how I rate rumors.
One is in line with AMD's statement and past behavior. It therefore has a higher "rating" than the other, believable rumor in line with AMD's statement but not in line with past behavior. I cannot explain it any more simply. It is rated by AMD's statements and past behavior. If you must reply to me again asking how to rate rumors I have to question if you are reading my posts at all. Stop. There is not a single question remaining about how I look at rumors.

Again, does not take AI craze into account. That is a new thing, so the past is N/A.

Regarding trust me bros, do you rate all equal so we can pull any random bro off the street and give you an estimate?

I think you rate them based on that you”d like your Zen6 in Q4 2026, so you choose the trust me bro that gives you the prediction you want to hear.

 

[OneEng2]

On the upside: We should stop seeing complaints about AMD "core stagnation" for a few generations?

LOL. Only until Intel releases a 52 core DT . I suspect there will be plenty of complaining. "If you give a mouse a cookie ....".

The question is if Zen6 DT now will be release in Q1 or Q2 2027.

Also, what happened to the dead certain trust me bro ~October 2026 predictions by some?

"Smoke on the water ..... fire in the sky ..... " .

I suspect, even if this rumor is true, Zen 6 will still launch in 2026Q4. Simply not for mere AM5 peons.
And, as usual, they are extrapolating far too much (performance, availability, potential pricing) from the server parts (which AMD actually cares about).

AMD has consistently stated their strategy is "Server First".

They care about DIY very much; they own the place and will collect hearthy margins there.

Yes, but the make much better margins in DC.

The slide is in the article poke01 linked. AMD's pretty vague. Zen 6. 2026.
If that includes Olympic Ridge or not is a mystery. But I see no reason to take this rumor seriously except that I already believe AMD doesn't prioritize desktop/DIY.

They don't prioritize desktop/DIY. I agree. They prioritize DC per their own statements. This isn't to say that they don't currently enjoy a dominant position in the desktop/DIY market, because they do. It's just a decreasing market AND there is absolutely no market pressure on ZEN 5 X3D ..... so what's the hurry?

That's how I currently envision it anyway.

 

[gdansk]

I think you rate them based on that you”d like your Zen6 in Q4 2026, so you choose the trust me bro that gives you the prediction you want to hear.

I'm waiting for X3D/NVL-SK comparison so it doesn't matter.
You're being a bit of a regurgitator here. I had mentioned AI & DRAM as early as... MANY posts ago:

At the end of the day Venice riding the AI boom is a better opportunity than prioritizing a market that can't afford DRAM.

But I can almost guarantee that the correct thing to do is weight rumors based on AMD's past behavior. 30 months between desktop launches would be Phenom-tier disaster from AMD. I'm not weighing DRAM shortages and AI any heavier than past behavior. It's complete speculation that it would impact their plan. Believable, but more speculative. Desktop has officially launched before server for a reason. And I doubt that reason has changed. That IOD and platform is simpler to validate.

 

[coercitiv]

I think you rate them based on that you”d like your Zen6 in Q4 2026

Do you REALLY wanna' do this?!

 

[adroc_thurston]

Yes, but the make much better margins in DC.

Not really, no.
More revenue though.

 

[Jan Olšan]

Is benchlife accurate?

https://videocardz.com/newz/amd-zen-6-desktop-ryzen-olympic-ridge-reportedly-set-to-launch-in-2027

I'm afraid so. It is possible that this isn't a very "strong" claim in the sense that it is partially their evaluation and not based on hard data/roadmaps/leaked launch-window info - in such case their estimate may turn out to be too pessimistic, perhaps? The text is not extremely clear on which sources they base the "only in 2027" conclusion on.

But if they actually say that based on knnowledge of internal plans, then RIP. My gut feeling is that we definitely should not ignore this, Benchlife is certainly no WCCFtech. Of course the news could turn out to be wrong, but probably not without a good reason for it.

Benchlife started as site of the guy who had good (IIRC) hardware leaks on the late VR-Zone Chinese cca 2015 and before - the very Chris L signed under this report, I think.
So they have been around, long experience and all, plus clearly do have contacts. I think they often leaked specs of umpcoming Intel desktop/HEDT processors.

And, they posted docs about AM4 platform like year and half in advance. The docs even referred to the socket as FM3, at that point.

最快 2017 年，AMD Zen 架構導入 APU 時間要再等等 - BenchLife.info

想要看到 AMD Zen 架構出現在 APU 平台中，你可能得等到 2017 年才有機會一瞥風采。 […]

benchlife.info

規格將全面翻新，AMD Summit Ridge 架構初步確定 - BenchLife.info

990FX 推出後，AMD 至今從未再更新自家晶片組，使得與 Intel 間的差距從 CPU 架構延 […]

benchlife.info

AMD 平台相容性小改，Bristol Ridge 延用 Excavator 架構 - BenchLife.info

想在 2016 年看到 AMD 重返榮耀的玩家得要失望了，Zen 架構將只會出現在 Summit R […]

benchlife.info

 

[gdansk]

I'm afraid so. It is possible that this isn't a very "strong" claim in the sense that it is partially their evaluation and not based on hard data/roadmaps/leaked launch-window info - in such case their estimate may turn out to be too pessimistic, perhaps? The text is not extremely clear on which sources they base the "only in 2027" conclusion on.

But if they actually say that based on knnowledge of internal plans, then RIP.

Benchlife started as site of the guy who had good (IIRC) hardware leaks on the late VR-Zone Chinese cca 2015 and before - the very Chris L signed under this report, I think.
So they have been around, long experience and all, plus clearly do have contacts. I think they often leaked specs of umpcoming Intel desktop/HEDT processors.

And, they posted docs about AM4 platform like year and half in advance. The docs even referred to the socket as FM3, at that point.

最快 2017 年，AMD Zen 架構導入 APU 時間要再等等 - BenchLife.info

想要看到 AMD Zen 架構出現在 APU 平台中，你可能得等到 2017 年才有機會一瞥風采。 […]

benchlife.info

規格將全面翻新，AMD Summit Ridge 架構初步確定 - BenchLife.info

990FX 推出後，AMD 至今從未再更新自家晶片組，使得與 Intel 間的差距從 CPU 架構延 […]

benchlife.info

AMD 平台相容性小改，Bristol Ridge 延用 Excavator 架構 - BenchLife.info

想在 2016 年看到 AMD 重返榮耀的玩家得要失望了，Zen 架構將只會出現在 Summit R […]

benchlife.info

The original article has a somewhat loose sense of time

AMD AM5 腳位從 2023 年的 Zen 4 處理器架構開始使用。

I wouldn't be surprised if the same flexibility of linear time occurs with Zen 6 DT.

 

[Jan Olšan]

Accurate? Who knows.
But it is very, very believable. Why give DIY their SKUs on time if they can sell them as Epyc instead?

You can be more aggressive with client launch wrt to going to market early, tuning and fixing AGESA glitches later.
The server probably holds on with launch or longer so that they can run all sorts of validations and tests for longer (while on the other hand, some customers do get early access, but with knowledge they work with samples, not final product). Desktop can even be seen as (beta) testing waters before server launch.

 

[Jan Olšan]

The original article has a somewhat loose sense of time

There's another thing - I think that the article's main topic isn't really leaking a Zen 6 delay. Just look at the title, I think from their point of view it is mainly a report on the SKU configurations leak by HXL.
The bit about launch being 2027 is addded there almost as a background, as if it was something that was generally known before, not anything new.

So there is a chance they don't base that on some newly gained internal information and in that case it could be inaccurate. But I may be reading too much into it, I'm not a chinese speaker or reader and can only work with the MTL slop rendition into english.

I wouldn't be surprised if the same flexibility of linear time occurs with Zen 6 DT.

 

[adroc_thurston]

Why give DIY their SKUs on time if they can sell them as Epyc instead?

EPYC is different packaging and different CCD.

 

[OneEng2]

But I can almost guarantee that the correct thing to do is weight rumors based on AMD's past behavior. 30 months between desktop launches would be Phenom-tier disaster from AMD.

The difference being that the limiting factor between generations ISN'T the new processor design. It's the new process design.

Additionally, due to the GREATLY reduced speed at which transistor density is improving, each new CPU architecture is more of a tweak than a redesign as they essentially only have 15-20% more transistor budget even AFTER 4 years of process improvements.

AMD (and ZEN 6) is going to be the last of the big bumps IMO. Transistor density improvements for N2 class GAA is going to go slow for quite some time (until High NA Lithography starts being used) IMO.

AMD is going to have to settle in for the long haul on the Zen 6 class architecture IMO. I am not expecting Zen 7 anytime soon, nor do I expect the same level of change from 6-7 as we should see from 5-6.

 

[yuri69]

We know the first wave of Zen 6 mobile SKUs launches in Jan 2027 at CES. Then rumors put DT Zen 6 to Oct 2026. So launching both lines together at CES wouldn't really change much...

 

[gdansk]

So launching both lines together at CES wouldn't really change much...

It would be the first time DT launches after server. Maybe the final "realization" of Zen server supremacy. They're usually shipping server parts to certain customers before its launch but now to make it official would probably be the beginning of the end of Zen dominance over DIY. Maybe not with NVL, but perhaps the part after that. I suspect NVL bLLC will be competitive with 9800X3D so to give Intel a few months of being on top in 1T, MT, and good enough in gaming would be hilarious self-sabotage. All AMD needs is a paper launch to avoid 90 "Intel is back" reviews.

 

[yuri69]

It would be the first time DT launches after server. Maybe the final "realization" of Zen server supremacy. They're usually shipping server parts to certain customers before its launch but now to make it official would probably be the beginning of the end of Zen for DIY. Maybe not with NVL, but perhaps the part after that. I suspect NVL bLLC will be competitive with 9800X3D so to give Intel a few months of being on top in 1T, MT, and good enough in gaming would be hilarious. All AMD needs is a paper launch to avoid 90 "Intel is back" reviews.

Some thoughts:

• As you said, server launches means very little since hyperscalers and OEMs have the chips available in advance of any official shiny launch presentation.
• Our resident "+40IPC person" repeatedly mentions desktop and server chips being way more different than in previous gens.
• Order is not given as the last gen was the first gen launching mobile SKUs first.
• AMD never misses an opportunity to miss an opportunity.
• The number of different tape outs seems to be even higher than last time - it possibly shakes things up.

AMD (and ZEN 6) is going to be the last of the big bumps IMO. Transistor density improvements for N2 class GAA is going to go slow for quite some time (until High NA Lithography starts being used) IMO.

AMD is going to have to settle in for the long haul on the Zen 6 class architecture IMO. I am not expecting Zen 7 anytime soon, nor do I expect the same level of change from 6-7 as we should see from 5-6.

Huh? Zen 6 does not seem to be a big bump - it is more like in Intel's territory of ~10% IPC.

I guess nobody expects Zen 7 "soon", as before 2028/2029.

 

[adroc_thurston]

Our resident "+40IPC person" repeatedly mentions desktop and server chips being way more different than in previous gens.

They're not really way more different.
Just different CCDs and packaging.

AMD never misses an opportunity to miss an opportunity.

reddit

AMD (and ZEN 6) is going to be the last of the big bumps IMO. Transistor density improvements for N2 class GAA is going to go slow for quite some time (until High NA Lithography starts being used) IMO.

A14 is a very good node and also an SRAM bitcell shrink. wtaf are you about.

AMD is going to have to settle in for the long haul on the Zen 6 class architecture IMO. I am not expecting Zen 7 anytime soon, nor do I expect the same level of change from 6-7 as we should see from 5-6.

Zen7 is q3'28.
more chungus too.

Additionally, due to the GREATLY reduced speed at which transistor density is improving, each new CPU architecture is more of a tweak than a redesign as they essentially only have 15-20% more transistor budget even AFTER 4 years of process improvements.

xtor budgets matter primarily for GPUs.

I suspect NVL bLLC will be competitive with 9800X3D

that would be a catastrophe.

 

[gdansk]

As you said, server launches means very little since hyperscalers and OEMs have the chips available in advance of any official shiny launch presentation.

Something Zen 6 should officially launch this year. Server is 99%. But I posit AMD won't change their process despite this one article claiming Zen 4 launched in 2023 also saying Zen 6 will launch in 2027.

Our resident "+40IPC person" repeatedly mentions desktop and server chips being way more different than in previous gens.

I think it's comparable to Zen 5. Separate CCDs but AMD put them in specific Granite Ridge SKUs anyway. Whatever differences there are don't seem to be that important.

Order is not given as the last gen was the first gen launching mobile SKUs first.

But we're pretty confident that isn't the case this time. But we're 5.5 out of 5 Zens officially launching on DT before server. Again, probably due to validation.

 

[adroc_thurston]

Again, probably due to validation.

It just goes in the TO order generally.

 

[itsmydamnation]

Hvabehar? What does that have to do with release dates on DT etc.

cast doubt on people saying delay because AI + DDR.

AI has alot of bottlenecks for supply raw wafer AFAIK isnt one of them.
AM5 having a big install base helps to ofset the DDR pricing.

 

[Joe NYC]

I wonder if AMD is going to run into supply issues in 2027 in regards to 2nm wafers.

Only if the AI customers all decide to abandon NVidia and switch to AMD Mi450.

 

[itsmydamnation]

Only if the AI customers all decide to abandon NVidia and switch to AMD Mi450.

i would assume many other supply chain issues before wafers , packaging , HBM etc

 

[adroc_thurston]

Only if the AI customers all decide to abandon NVidia and switch to AMD Mi450.

The supply chain for those racks is extremely complex.
Just don't think about it too hard.

 

[Joe NYC]

We know the first wave of Zen 6 mobile SKUs launches in Jan 2027 at CES. Then rumors put DT Zen 6 to Oct 2026. So launching both lines together at CES wouldn't really change much...

It would be like a replay of disastrous rollout of 5800x3d, releasing it first as Milan-X, which no one was even expecting, and postponing desktop 3-4 months, handing Intel undisputed performance leadership during that period.

It was a mistake results of which that took 2-4 years to overcome.

Will AMD repeat it? Certainly possible.

 

[adroc_thurston]

It would be like a replay of disastrous rollout of 5800x3d, releasing it first as Milan-X, which no one was even expecting, and postponing desktop 3-4 months, handing Intel undisputed performance leadership during that period.

It was a mistake results of which that took 2-4 years to overcome.

Will AMD repeat it? Certainly possible.

wut?
2-4 years to what