Question Zen 6 Speculation Thread

[soresu]

I switched to CachyOS a while ago.

There are multiple projects aiming to streamline a containerized Windows install for apps and games not compatible with wine. Does not work with anticheat in games, however. The containers let you run things like adobe and office natively.

Apparently there is a way to hide a virtualized windows installation from anticheat. Understandably, you won’t find guides that show how to do this easily available. A solution like that would theoretically give you 100% compatibility, however.

The Linux desktop inches forward.

My assumption is that virtualised OS instances run in hypervisors that work in a similar fashion to the hacking methods the anticheat is designed to detect and counter.

Possibly the solution consists of turning off certain hypervisor features that are otherwise beneficial for administration or maintenance of the virtualised containers by which these anti cheat routines can detect their operation - sort of like it's acting as a fake client for the hypervisor saying "I'm here, virtualise me!" but the hypervisor isn't responding to it anymore.

 

[OneEng2]

Stuff that doesn't apply to Zen6?

Zen 5 is completely socket power limited in highly MT applications.

What math are you using where 50% more cores will not be limited by something other than clock speed for Zen 6?

We can all "click our heals together" and say "There's no processor like Zen 6" over and over again, but it won't change physics.

 

[adroc_thurston]

Zen 5 is completely socket power limited in highly MT applications.

Everything in the whole wide world is socket power limited in hightly nT applications.

What math are you using where 50% more cores will not be limited by something other than clock speed for Zen 6?

the math where you get 60% lower power iso freq by going to N2p.

We can all "click our heals together" and say "There's no processor like Zen 6" over and over again, but it won't change physics.

Indeed there ain't!
That's why AMD rules server market with an iron fist.

 

[Fjodor2001]

NPU

NA

XDNA3 80-110 TOPS [for Zen 6 / Gator Range IOD]

My analysis

Thanks for the info. Just wondering what this analysis is based on?

 

[adroc_thurston]

Just wondering what this analysis is based on?

him hitting datboi crackpipe

 

[jpiniero]

Indeed there ain't!
That's why AMD rules server market with an iron fist.

Now that's kind of a stretch. There's still plenty of people still buying Intel, even though Intel server processors suck.

 

[adroc_thurston]

Now that's kind of a stretch

nope.

There's still plenty of people still buying Intel, even though Intel server processors suck.

Legacy customers do not define hyperscaler roadmaps and general competitive positioning.
with DMR-SP's demise even they are on borrowed time, anyway.

 

[OneEng2]

the math where you get 60% lower power iso freq by going to N2p.

Intel is also going to get a big boost moving to NVL to N2. There will be no process advantage between the 2 companies on this level (which will be really interesting for analysis).

Still, I went back and did some math and a little research. Looks like Zen 5 9950x is throttling when running CB2024 (as we have discussed extensively here with @Hulk). I also looked back at the power efficiency of 9950x vs 285K.

As much as it pains me to admit, you might be correct .... but it will be close in heavy FP and high ILP loads like CB24. Zen 6 could win by lieu of its better power efficiency in high MT workloads.

Now, lets review this discussion somewhere in H2 2026 .

 

[itsmydamnation]

power eff is a funny one , if AMD could volt and scale clock like intel P cores do im sure there would be a PBO mode that would do it.

 

[Hulk]

Intel is also going to get a big boost moving to NVL to N2. There will be no process advantage between the 2 companies on this level (which will be really interesting for analysis).

Still, I went back and did some math and a little research. Looks like Zen 5 9950x is throttling when running CB2024 (as we have discussed extensively here with @Hulk). I also looked back at the power efficiency of 9950x vs 285K.

As much as it pains me to admit, you might be correct .... but it will be close in heavy FP and high ILP loads like CB24. Zen 6 could win by lieu of its better power efficiency in high MT workloads.

Now, lets review this discussion somewhere in H2 2026 .

Yes. I'm not as concerned with max frequency of the new node when it comes to boosting a few cores as I am with hammering all cores and maintaining a decent frequency and lowish (compared to Zen 5) power usage.

 

[inquiss]

Yes. I'm not as concerned with max frequency of the new node when it comes to boosting a few cores as I am with hammering all cores and maintaining a decent frequency and lowish (compared to Zen 5) power usage.

On desktop any power that's available for highly threaded workloads will go into boosting the cores for those workloads. Expecting frequency, good, expecting less power usage when the workload needs that power..won't be happening imo. Why would they leave that on the table?

 

[eek2121]

Intel is also going to get a big boost moving to NVL to N2. There will be no process advantage between the 2 companies on this level (which will be really interesting for analysis).

Except Intel’s P-cores are inefficient and their E-cores aren’t designed for high clocks.

 

[Covfefe]

power eff is a funny one , if AMD could volt and scale clock like intel P cores do im sure there would be a PBO mode that would do it.

View attachment 134642

Intel lost its clock speed advantage when they moved on from the Intel 7 node. Zen5 clocks as high as Lion Cove.

Spoiler: TPU Clock Frequency Graphs

 

[inquiss]

Intel lost its clock speed advantage when they moved on from the Intel 7 node. Zen5 clocks as high as Lion Cove.

Spoiler: TPU Clock Frequency Graphs

View attachment 134660

View attachment 134661

Really interesting and helpful graphs

 

[OneEng2]

Except Intel’s P-cores are inefficient and their E-cores aren’t designed for high clocks.

And with Zen 6 and NVL we will have definitive proof one way or the other which DESIGN is better since the process node will level the playing field.

 

[Thibsie]

There’s still a chance we’ll get Win 12 before October 2026. That’s what I’m hoping for, and then I’ll upgrade to Zen6 or NVL-S at the same time as switching to Win 12, assuming those CPUs will be released latest a few months after that deadline.

I can probably survive a few months without security updates. But if Win 12 will be released a lot later than that I’ll be in trouble and will have to decide on a new plan. Worst case I’ll have to jump to Win 11 instead on the new PC and stay on that until Win 12 is released.

Long live win10 LTSC

 

[Fjodor2001]

Long live win10 LTSC

Which variant of that do you have in mind?

Win 10 Enterprise is EOL January 2027. There is Win 10 IoT as well with EOL January 2032, but I don’t think many use that on desktop/laptop. Also the former is Volume licensing only, and the latter Volume and OEM licensing. So not very common for private use, although I guess there are ways of obtaining it also for private use.

Also, there of course other drawbacks like no feature upgrades.

 

[Timorous]

Also, there of course other drawbacks like no feature upgrades.

Given the way MS is taking windows that is a positive not a negative.

 

[Fjodor2001]

Given the way MS is taking windows that is a positive not a negative.

Yeah, I agree. But those feature upgrades could also include later versions of libraries etc which may be needed/beneficial for Apps.

 

[511]

Stuff that doesn't apply to Zen6?

well have you seen the OEM/Board Vendor skill issues that make everything possible?

 

[Joe NYC]

AMD inadvertently posted info on embedded Venice on AMD web site, then took it down.

96 cores = 8 x 12 core CCDs

https://twitter.com/x/status/1996157021335507100

 

[Tarkin77]

Yesterday at UBS’s 2025 Global Technology Conference, Lisa said :"We're super excited about our next-generation Venice CPUs. We think that extends our leadership going forward, and that extends as we go into the next generation as well"

 

[Geddagod]

Yesterday at UBS’s 2025 Global Technology Conference, Lisa said :"We're super excited about our next-generation Venice CPUs. We think that extends our leadership going forward, and that extends as we go into the next generation as well"

Dunno if this means verano or if it means Zen 7 Epyc, but yea even Intel is admitting they won't be competitive in servers till 2028 at best.

 

[Joe NYC]

Dunno if this means verano or if it means Zen 7 Epyc, but yea even Intel is admitting they won't be competitive in servers till 2028 at best.

Yeah, I think she said "next next generation", which implies Zen 7

 

[511]

Dunno if this means verano or if it means Zen 7 Epyc, but yea even Intel is admitting they won't be competitive in servers till 2028 at best.

Likely Zen 7