Question Zen 6 Speculation Thread

[adroc_thurston]

how does AMD/TSMC determine which CCDs are "Leaky" and unsuitable for EPYC usage

There is no EPYC. Turin-X got shelved.

Do they batch all the CCDs from each wafer, drop one into a test substrate, test that single CCD, then bin the rest of the batch based on that single result?

It's normal dicing and test; just happens *post*-bonding, making it a separate binning pool not shared with normal non-X3D CCDs.

 

[Hulk]

No it's a normal derived core.

It would be nice of you to let Tom's know so they can update their article.

"As it turns out, Zen 6 is not exactly an evolution of Zen 5, but rather an all-new design with a different ideology."
- Toms Harware

 

[adroc_thurston]

It would be nice of you to let Tom's know so they can update their article.

Tom's is like one step removed from wccftech at this point so uhhh

 

[Hulk]

How many execution ports and how many execution units for Zen 5 do you count from this block diagram?

 

[OneEng2]

I'm not convinced this really matters outside the DIY market where people are upgrading piecemeal. No one is upgrading CPUs of PCs they bought at Dell, whether it was a consumer buying one at a time or a corporation buying them in lots of 10K at a time.

Heck, we're well under 10% of fully built PCs that EVER have their case opened post sale, even for simple upgrades/repairs like RAM or storage. So you can bet the number that have their CPU upgraded is likely to be under 1%.

Do PC OEMs care? Does it make life harder for Dell if every new Intel CPU has a different socket? I have a hard time imagining it affects them. Same for any reasonable large white box OEM. Do the small shops that build a few hundred to a few thousand PCs a year still exist? Maybe they care.

Beyond them, it is just the DIY market who cares, and even then not people like me who build a PC like the one I built last May and won't later upgrade the CPU - I'll just build a new PC with new everything when I reach that point (and by that time we'll be on DDR6 or LPDDR6 LPCAMMs, heck maybe even LPDDR7 lol) I've never seen any stats on how big that is, i.e. if the PC market as a whole is 250 million a year, what percentage of that is DIY? 1%? 5%? I really don't know.

I agree. The focus on gaming in this forum with respect to its importance in the overall CPU strategy of a company is very outsized compared to reality.

Intel stated in a presentation about 10 years ago or so the enthusiasts have an outwardly impact compared to average users and concluded they affect the buying decisions of many. I can't remember the number but it was close to ten.

Yes, to some degree I believe this to be true; however, I believe that this effect is decreasing rapidly as desktops are replaced by laptops and even phones and tablets..... which are disposable in comparison to desktop.

The impact is not zero. Whether Intel's way of supporting barely 2 years matter versus AMD's of 6+ years is going to be an everlasting war however.

Agree. It isn't 0. I just think that other factors are more important. What would be interesting is to understand the indirect impacts of the platform longevity (cost effective development of multiple generations of cpus and mb's?).

They are releasing it because there always will be some clueless guy who thinks he needs more cores

I think that there will be way more clueless guys than knowledgeable ones. It might be a good bet. People is stupid .

Intel is doing it because brute-force MT is their only chance to win any benchmarks.

Yes. Plus 48 > 24. Even people that is stupid can understand that.

Based on the article below Zen 6 is going to be more revolutionary than evolutionary. I have a feeling they are starting with dispatch and execution of Zen 5 and reworking everything around that .

That would make it a very interesting release .... especially for we geeks . I hope Tom's article is correct!

- removing SMT was dumb, especially from a marketing perspective

Yes, and from an architecture POV IMO. 5% transistor budget increase for 20% MT improvement in DC with no discernable increase in power usage.

 

[StefanR5R]

I hope Tom's article is correct!

Tom's article is bull excrement. It was written as if Zen 5 doesn't exist. (Its last paragraph was written as if Zen doesn't exist.)

 

[adroc_thurston]

Plus 48 > 24. Even people that is stupid can understand that.

I mean 8 > 4 but Sandy desktop still stomped all over the poorpoor Bulldozer. much to thingk about

 

[Hulk]

I mean 8 > 4 but Sandy desktop still stomped all over the poorpoor Bulldozer. much to thingk about

Bulldozer had some inherent flaws that weren't corrected until around Steamroller. Current Zen and Intel P/E cores aren't inherently flawed today.

 

[Cheesecake16]

How many execution ports and how many execution units for Zen 5 do you count from this block diagram?View attachment 137047

I count 16 ports...

 

[Joe NYC]

Doesn't need 'em.

If M4/M5 has 16 MB L2, shared between 4 cores, in single ST, in theory, the single core has access to 16 MB. As a result, Apple M processors look very good in ST, but then, lose some of that edge in MT.

 

[adroc_thurston]

If M4/M5 has 16 MB L2, shared between 4 cores, in single ST, in theory, the single core has access to 16 MB.

It's a good caching scheme for low CC client parts.

As a result, Apple M processors look very good in ST, but then, lose some of that edge in MT.

No. But shared L2's are hardly scalable beyond 6c clusters.

 

[luro]

People here act like Intel’s current generation doesn’t already have more cores, yet AMD still outsells them in every price tier

edit: a gazillion cores won't save Intel if they don't have the ST crown

 

[jpiniero]

People here act like Intel’s current generation doesn’t already have more cores, yet AMD still outsells them in every price tier

edit: a gazillion cores won't save Intel if they don't have the ST crown

I'm not sure what that dude's issue is, but people who would actually want more corez would want real cores and not the Cinebench accelerators.

 

[Hulk]

I'm not sure what that dude's issue is, but people who would actually want more corez would want real cores and not the Cinebench accelerators.

I used to like having Cinebench accelerators when CB did the swarm of bees thing because it was fun too watch. I don't care for it now.

 

[gdansk]

I used to like having Cinebench accelerators when CB did the swarm of bees thing because it was fun too watch. I don't care for it now.

If we go by worker bee visualization when running CB2023 AMD is ahead right now (32 Z5 workers vs 24 ARL). It'll be equal again (or near about) for Z6 vs NVL.

But don't be offended when I say this amusing metric probably doesn't impact the purchasing behavior of many.

 

[Fjodor2001]

I'm not sure what that dude's issue is, but people who would actually want more corez would want real cores and not the Cinebench accelerators.

It’s perf/thread that is the interesting metric in this case if anything.

And if that’s what you’re aiming for then AMD should drop SMT. Because one Zen6 SMT thread will have lower perf than one NVL-S E-Core thread, assuming 48T maxed out on both CPUs.

 

[gdansk]

And if that’s what you’re aiming for then AMD should drop SMT.

Why? It doesn't make ST perf lower. It only lowers stupid metrics no one cares about, like Cinebench points per thread.

 

[AMDK11]

How many execution ports and how many execution units for Zen 5 do you count from this block diagram?View attachment 137047

Exactly 16 ports and execution units.

 

[inquiss]

It’s perf/thread that is the interesting metric in this case if anything.

And if that’s what you’re aiming for then AMD should drop SMT. Because one Zen6 SMT thread will have lower perf than one NVL-S E-Core thread, assuming 48T maxed out on both CPUs.

No it won't, please stop.

 

[Hulk]

If we go by worker bee visualization when running CB2023 AMD is ahead right now (32 Z5 workers vs 24 ARL). It'll be equal again (or near about) for Z6 vs NVL.

But don't be offended when I say this amusing metric probably doesn't impact the purchasing behavior of many.

For all of our bickering AMD and Intel have been quite competetive for the last few generations outside of gaming performance, which honestly is more GPU bound than CPU until you get to extremely high frame rates. Jeez, I remember the days when anything over 30fps was good and 60fps was amazing.

I agree that Zen 6 and Nova Lake will be competitive with one another. For me I'm focused on two things. First, who has the better ST performance? Second, who can maintain high clocks under high MT loads without going nuclear and requiring a nuke to power it?

 

[adroc_thurston]

For all of our bickering AMD and Intel have been quite competetive for the last few generations outside of gaming performance

"quite competitive outside of DIY thing that Actually Matters"

I agree that Zen 6 and Nova Lake will be competitive with one another.

nope

For me I'm focused on two things. First, who has the better ST performance? Second, who can maintain high clocks under high MT loads without going nuclear and requiring a nuke to power it?

you already know the answer.

 

[Hulk]

"quite competitive outside of DIY thing that Actually Matters"

nope

you already know the answer.

I honestly have no idea what you are trying to communicate. If you want to participate you need to put a little more effort into the keys.

 

[adroc_thurston]

I honestly have no idea what you are trying to communicate

You should try reading.

 

[ETI4711]

It would be nice of you to let Tom's know so they can update their article.

"As it turns out, Zen 6 is not exactly an evolution of Zen 5, but rather an all-new design with a different ideology."
- Toms Harware

A detailed response to this article:

【大原雄介の半導体業界こぼれ話】 次期CPU「Zen 6=大変革」説に物申す。AMD公式文書を冷静に読み解いてみた

12月12日、AMDは技術文書をアップデートし、「Performance Monitor Counters for AMD Family 1Ah Model 50h-57h Processors」を公開した。ちなみにこれに最初に気が付いたのはInstLatX64である。

pc.watch.impress.co.jp

 

[OneEng2]

People here act like Intel’s current generation doesn’t already have more cores, yet AMD still outsells them in every price tier

edit: a gazillion cores won't save Intel if they don't have the ST crown

For all of our bickering AMD and Intel have been quite competetive for the last few generations outside of gaming performance, which honestly is more GPU bound than CPU until you get to extremely high frame rates. Jeez, I remember the days when anything over 30fps was good and 60fps was amazing.

I agree that Zen 6 and Nova Lake will be competitive with one another. For me I'm focused on two things. First, who has the better ST performance? Second, who can maintain high clocks under high MT loads without going nuclear and requiring a nuke to power it?

I have a feeling you are right in most respects. I think that they will each have their favorite benchies. Intel will dominate CB and it is likely that AMD will still dominate (or at least have a bit of wiggle room) in gaming. Everything else I think might be close.

I honestly have no idea what you are trying to communicate. If you want to participate you need to put a little more effort into the keys.

"I have a dream" Dr. Martin Luther King Jr. August 28, 1963