Question Zen 6 Speculation Thread

[branch_suggestion]

Dr. Lisa Su, Chair and CEO of AMD, to Keynote CES 2026 on how AI is Changing the World

www.ces.tech

What could Lisa be cooking?

 

[adroc_thurston]

Dr. Lisa Su, Chair and CEO of AMD, to Keynote CES 2026 on how AI is Changing the World

www.ces.tech

What could Lisa be cooking?

Roadmaps

 

[branch_suggestion]

Roadmaps

It's gonna reiterate FAD isn't it.
Still they could do demos of some parts by then.

 

[adroc_thurston]

It's gonna reiterate FAD isn't it.

Duh.
It's Lisa.

Still they could do demos of some parts by then.

Obviously, Zen6's been doing well for a while.

 

[Thunder 57]

Dr. Lisa Su, Chair and CEO of AMD, to Keynote CES 2026 on how AI is Changing the World

www.ces.tech

What could Lisa be cooking?

CES? Meh that's far off and CES is usually boring. Hot Chips is usually good but was a let down this year.

 

[gdansk]

Hot Chips is usually good but was a let down this year.

You will not see Zen 6 at HC until August 2027.
If they sneak in for 2026, it might be before the launch.

 

[adroc_thurston]

Hot Chips is usually good but was a let down this year.

AMD HotChips presentations are nearly always rewarmed marketing .pptware.
They're not like Apple bad, but still bad wrt interesting technical disclosures.

 

[Thunder 57]

You will not see Zen 6 at HC until August 2027.
If they sneak in for 2026, it might be before the launch.

I thought Hot Chips was over?

AMD HotChips presentations are nearly always rewarmed marketing .pptware.
They're not like Apple bad, but still bad wrt interesting technical disclosures.

IIRC there was some real good info about the OG Zen at Hot Chips including the first cinebench score.

 

[adroc_thurston]

IIRC there was some real good info about the OG Zen at Hot Chips including the first cinebench score.

That was before AMD got real.

 

[Thunder 57]

That was before AMD got real.

Got real? I don't understand. Please explain in more than 2-4 words.

 

[adroc_thurston]

Got real? I don't understand. Please explain in more than 2-4 words.

They were broke losers and needed all the attention they could get.
Now they don't.

 

[Thunder 57]

They were broke losers and needed all the attention they could get.
Now they don't.

That makes perfect sense and I appreciate the response rather than just "Got real".

 

[OneEng2]

I would say removing it and not having anything to compensate is a mistake. If they took the classes in magical frontends at school of hardware wizardy (I am afraid the teachers went to work at Apple so the classes are on hold), then getting rid of SMT wouldn't be so bad. Alas, they did not and are behind both Apple and AMD.

I think that SMT is a very good use of transistors .... for DC. I actually believe that in desktop and laptop, focusing the same 5% transistor budget in other places pays off more frequently.... but if you are talking about a business strategy, and don't want 2 completely different core technologies, SMT is the best bet IMO.

That makes perfect sense and I appreciate the response rather than just "Got real".

Ya know, I really wanted to like this post more than once but it wont let me .

 

[soresu]

Even compared to Yonah.

Oh ye damn I forgot about Yonah, that would be the most direct µArch predecessor.

Its hi end SKU is literally even called Core Duo.

 

[adroc_thurston]

Oh ye damn I forgot about Yonah

Even Intel did.
It was like the most low-key Intel release in ever.
All the IDF shilling and every roadmap pointed directly at Merom/Conroe. It was funny.

 

[MS_AT]

I actually believe that in desktop and laptop, focusing the same 5% transistor budget in other places pays off more frequently....

Well, do you have some examples? For Intel it did not exactly work out and I guess the only reason they cut their SMT was to save of verification time since they are behind the schedule all the time. Since the client machine is usually doing lots of low IPC stuff then SMT helps to smooth things over giving you a chance to use a silicon that would be otherwise be left unused. Apple solves this problem without SMT but it seems they have much more sophisticated front-end others are unable to reproduce.

Therefore I am not sure how AMD could spend those 5% silicon area better than continue to use SMT. I mean I would expect that whatever Apple does requires more than 5% if we even can reason in this way as I guess its deeply ingrained in the core so cannot be singled out.

 

[Joe NYC]

Roadmaps

CES is more likely products. I am not quite sure what products might be launching - that are going to be more consequential then last CES, which Lisa skipped (Strix Halo, Kraken, 9700 XT).

Gorgon Point would not be worthwhile, if it is just a refresh and it maybe too early for either Zen 6 or Mi400

 

[511]

Why is AMD Launching MI400 it's CES

 

[Joe NYC]

Why is AMD Launching MI400 it's CES

I don't think AMD is launching Mi400 at CES.

 

[marees]

Why is AMD Launching MI400 it's CES

MI400 launch is June (Q2) 2026

 

[yuri69]

CES is more likely products. I am not quite sure what products might be launching - that are going to be more consequential then last CES, which Lisa skipped (Strix Halo, Kraken, 9700 XT).

Gorgon Point would not be worthwhile, if it is just a refresh and it maybe too early for either Zen 6 or Mi400

January 2026 sounds like the launch of those *great* AI APU refreshes. Grogon Point with premium AI Copilot TOPS+++ and of course a Kraken refresh with a bit less permium AI TOPS+. AI AI AI

 

[MadRat]

People honestly do not care about AI workloads. If anything not many welcome the technology.

 

[igor_kavinski]

People honestly do not care about AI workloads. If anything not many welcome the technology.

That's because many have not tried it. I hate the online stuff because they make you pay per token used but the offline one is pretty great.

 

[Thunder 57]

That's because many have not tried it. I hate the online stuff because they make you pay per token used but the offline one is pretty great.

Call me a n00b but what offline one and what hardware do you need to run it?

 

[igor_kavinski]

Call me a n00b but what offline one and what hardware do you need to run it?

The easiest is LM Studio.

If you have at least a 5800X with 32GB DDR4-3200 (though 3600 or 3733 would be best), try it with the model gpt-oss-20b. May work with 16GB RAM but you will get very close to running out of RAM. It could work fine with slower CPUs but the wait time will be harder to deal with.

Ask it questions you are curious about. Give it a text or PDF document and ask questions about that. You will be surprised how useful it can be.