Question Zen 6 Speculation Thread

[adroc_thurston]

Even if desktop is a total shit show once again, which again, I doubt it will be as bad as it is with ARL vs Zen 5 X3D because of bLLC being a thing,

No, it'll be much worse.
At least now Intel has a node and pkg advantage.

and I still doubt Intel's design side won't at least survive.

They survive until Unified Core. If it sucks, IPG has to exit.

is probably referring to the whole desktop segment, OEM boxes make the most of it and Intel has plenty of momentum left there

This is the reason why Intel GM is 35%. That's not good.

on the mobile side AMD is giving them a break because they're too focused on AI.

Bull, they lost share last Q.

 

[Momoka_]

We're about to get a next-gen console from Sony with only 6 cores where only 4 of them are available to games.
Thanks to that, it does look like core stagnation in games might get extended for another decade or so.

According to recent rumors, PS6 will feature 8x Zen6c, while Xbox Next is opting for a hybrid 3x Zen6 + 8x zen6c setup. Although console CPUs are significantly weaker than desktop CPUs due to cache, frequency, and process node constraints, their performance is still enough for consoles thanks to UMA.

 

[maddie]

There's been reports saying if the AI bubble doesn’t pop or deflate, leaving RAM allocation free for other stuff, many consumer electronics' brands are going bankrupt, and then a bunch of retailers will follow. Unless there’s some form of government intervention.



We're about to get a next-gen console from Sony with only 6 cores where only 4 of them are available to games.
Thanks to that, it does look like core stagnation in games might get extended for another decade or so.

Rock meets hard place.

Ignoring the stock markets, most present US Capex is either AI or AI related, such as energy and physical infrastructure expansion. Not a good position.

 

[adroc_thurston]

plus medusa premium comes with 4p + 8c + 2lp config

It doesn't.

 

[OneEng2]

AFAIK Zen 6 makes use of the advanced packaging seen for Strix Halo, so it should also see significant improvements in latency and power. A great recap of the packaging used in Halo can be found here.

Thanks for that, and good point.

It will depend on all core clock rates which can differ from f-max quite a lot.

IPC uplifts seem to be similar ish but it seems Z6 will have a clock bump nvl won't. How that shakes out for all cores will determine the MT winner in the highly threaded parts.

Feeding all those cores may come at a significant power cost so it will also depend on if Intel are willing to juice it just to win a few benches.

AFAIK, Intel's "mont" cores currently appear to provide higher performance per watt over Zen 5. I think it will come down to socket power and PPW. Still seems likely to me that Intel will win this handily against a 24c Zen 6. Latest rumor is that the 52c NVL will consume 700W..... so Intel is indeed pushing higher

There could be as much as a 30%-40% all core clock gap between the two chips if things play out well for Zen 6. AMD is going from N4 to N2, which is a pretty big shrink. Add to that a 10-15% IPC increase and Intel has a pretty big hill to climb. While Intel is juicing PL2 for their next chip, They have a lot more cores to feed.

Possibly so.

W.r.t. price, for reference Zen5 TR launch price was $2499 (9970X, 32C) and $4999 (9980X, 64C).

OUCH!

I feel like Zen 6 will be a make-or-break moment for Intel. If they can't find a way to differentiate/compete, they (the chip design portion, at least) will be demolished, and that happens regardless of who has performance leadership, unless Intel can redefine performance and value as we know it, which likely isn't going to happen.

Intel has plenty of ways to "survive". I don't think Zen 6 will kill them, but it will make matters worse IMO.

Commercial buys a lot of laptop and laptop they're gonna continue bleeding share for a wide variety of reasons.

Desktop is currently the segment "bleeding share". AMD betting on "Server First" is why they are "winning" financially. X3D has made them the darling of gamers and the lucrative part of desktop, so they will certainly keep doing this .... but the future is DC. They know that.

Frankly, I think you guys are placing way too much importance on desktop. Even if desktop is a total shit show once again, which again, I doubt it will be as bad as it is with ARL vs Zen 5 X3D because of bLLC being a thing, Intel has a ton of momentum on DT, and the AI boom has reached servers, meaning even Intel's worse server CPUs will still at least sell, even if not for as high margins/prices as AMD's stuff.

Zen 6 can be amazing, and I still doubt Intel's design side won't at least survive.

Ding ding ding!!!

The only thing that can break them right now is failing on the manufacturing side, and we really don't want them to do that. I'd rather play with knives than bet on a positive outcome for consumers after an IFS collapse.

It's really a financial problem. Intel has shored up its weaknesses in the past by throwing away money to prop up market share. They could ONLY do this because they were making SO much money on DC and had a near monopoly on the x86 ecosystem (they sold processors in a package deal with lots of other chips needed on the MB).

Intel's insistance on maintaining process superiority when NOT commanding even a fraction of the chip market share that TSMC commands is (IMO) their biggest liability.

It's the money running out that will ultimately "kill Intel". AMD is simply running a MUCH more cost efficient operation. If Intel can fix this part, everything else will fall into place in the long run. If they can't? They will likely get sold .... unless the US government steps in to hold them above water.

 

[adroc_thurston]

AFAIK Zen 6 makes use of the advanced packaging seen for Strix Halo

Different, it's LSI.
Clients parts are not at TSMC either.

 

[adroc_thurston]

Intel's "mont" cores currently appear to provide higher performance per watt over Zen 5.

Which is why Atom xeons are megadead

 

[inquiss]

AFAIK, Intel's "mont" cores currently appear to provide higher performance per watt over Zen 5. I think it will come down to socket power and PPW. Still seems likely to me that Intel will win this handily against a 24c Zen 6. Latest rumor is that the 52c NVL will consume 700W..... so Intel is indeed pushing higher

Whether intel wins or not with a 52c Vs 24c in multithread is entirely irrelevant to either companies financial future. The market there is tiny. Should intel win there, it won't be relevant if they don't compete at normal core counts that people actually buy

 

[Kaffeekenan]

Different, it's LSI.
Clients parts are not at TSMC either.

Sooooo, is Zen 6 better or just different?

 

[adroc_thurston]

Sooooo, is Zen 6 better or just different?

Both!
You get a finer pitch (25um) with much lower insertion loss.
It's also packaged at ASE and not TSMC.

 

[ToTTenTranz]

According to recent rumors, PS6 will feature 8x Zen6c, while Xbox Next is opting for a hybrid 3x Zen6 + 8x zen6c setup.

Canis, the PS6 Handheld, is 4c Zen6 + 2c Zen6c (OS & Sony apps only).

you mean handhelds? they have a 15 watt budget. imagine playing PS5 games at that power budget

Sony already deployed "Power Saver Mode" on the PS5, and it's widely thought that this mode will allow the PS6 handheld to run PS5 games natively. It halves memory bandwidth, cuts down available CPU cores/threads to 4c/8t and fixes GPU clocks to "base clock" (probably around 1.8GHz if the AMD Github leak is any indication).

 

[Joe NYC]

I feel the need to drop this HBU video here:

I am not their biggest fan, however their coverage of hardware hasn't been terrible IMO (I am just a nerd who wants nerd details) yet I think they nailed yet another reason why Intel will struggle with Zen 6 competition, and, more dramatically, why Zen 7 may turn Intel into a Fab only business (that last part is purely speculation on my part, however) I don't see how Intel comes back from this with storage/memory shortages, no performance to bring to the table, and no socket longevity. I realistically don't see how they can even compete with Zen 6. Even the most hardcore Intel folks I know have adopted AMD CPUs/boards at this point.

Funny enough, IMO, if they don't survive this, it will be the constant platform changes that killed them, not the poor upgrades themselves.

I feel like Zen 6 will be a make-or-break moment for Intel. If they can't find a way to differentiate/compete, they (the chip design portion, at least) will be demolished, and that happens regardless of who has performance leadership, unless Intel can redefine performance and value as we know it, which likely isn't going to happen.

Just my 2 cents. Looking forward to seeing how things play out.

Good video, but I posted this comment:

Couple of nitpicks:
1) DDR5 revisionist history. DDR5 was a "benchmark special". DDR5-7200 did not exist at Alder Lake launch. High end DDR5 was $400+ vs $150 for DDR4.It is highly misleading to show results with DDR5-7200 which did not exist rather than DDR4 which overwhelming majority of Alder Lake buyers bought
2) With same information as today, in a bizzarro appearance on MLID Steve said: ""platform longevity means nothing". Is that what the title of the video is referring to? As far as being wrong?

 

[Joe NYC]

Here are more complete quotes from Steve. At the time he made them, they stuck in my head. I was thinking: "What is Steve smoking":

1. "platform longevity means nothing if the platform barely progresses over a 5year period it's like you bought in day one you know like if you bought a 7800 x3d and a 9800 x3d is 10% faster you that upgrade is not happening is it yeah and then if if Zen six is another 10% it's like ah you know I could upgrade my CPU for 20% Which is a reasonable performance gain on the CPU side or I could just hold on to this thing and wait for the next platform."

2. "AMD's platform longevity broad platform compatibility whatever you want to call it that's sort of got out the window for me it's no longer something that I'll be really talking about for this matchup."

 

[adroc_thurston]

Here are more complete quotes from Steve. At the time he made them, they stuck in my head. I was thinking: "What is Steve smoking":

1. "platform longevity means nothing if the platform barely progresses over a 5year period it's like you bought in day one you know like if you bought a 7800 x3d and a 9800 x3d is 10% faster you that upgrade is not happening is it yeah and then if if Zen six is another 10% it's like ah you know I could upgrade my CPU for 20% Which is a reasonable performance gain on the CPU side or I could just hold on to this thing and wait for the next platform."

2. "AMD's platform longevity broad platform compatibility whatever you want to call it that's sort of got out the window for me it's no longer something that I'll be really talking about for this matchup."

It's correct.
If your perf CAGR is 5% then your socket compat is worthless.

Yes. OUCH! . So do you really think 52C NVL-S will cost $2499-$4999 like 32-64C TR which you compared it to? If not, those TR models will be in trouble.

No one cares about cinememe machines.
Try thinking hard about 7.2G Zen6 or something.

 

[Joe NYC]

It's correct.
If your perf CAGR is 5% then your socket compat is worthless.

But the overall performance gain of AM4 Socket was 124% over the life of the socket makes a CAGR of 17.3%

That information was available to Steve when he made the comments that "platform longevity means nothing".

He just succumbed to Zen 5% mass hysteria, and he was not thinking clearly.

 

[Kryohi]

Try thinking hard about 7.2G Zen6 or something.

Wat

 

[adroc_thurston]

But the overall performance gain of AM4 Socket was 124% over the life of the socket makes a CAGR of 17.3%

That is precisely my point.
Socket compat only matters if your perf CAGR is strong.

Wat

It's the very very top of the range it could be at (26% fmax across a tock + 2 nodes).
Just some mildly funny things to imagine.

 

[Geddagod]

what

I'm tweaking, I meant ton of momentum on mobile lol

@Geddagod is probably referring to the whole desktop segment, OEM boxes make the most of it and Intel has plenty of momentum left there

No I'm just stupid and misspoke lol 😛

No, it'll be much worse.
At least now Intel has a node and pkg advantage.

I expect perf to be closer, and yea rn Intel has a cost disadvantage due to using a more advanced node and more advanced packaging.
Though I haven't seen any estimations to see if the manufacturing cost gap shrinks between ARL and Zen 5, vs NVL and Zen 6, due to how large the bLLC tiles are...

 

[gdansk]

Operating frequency of Venice is up a lot.
But I doubt that Olympic Ridge is up as much for a variety of reasons.

 

[adroc_thurston]

I expect perf to be closer

More far apart than ever.

Though I haven't seen any estimations to see if the manufacturing cost gap shrinks between ARL and Zen 5, vs NVL and Zen 6, due to how large the bLLC tiles are...

It's not about cost, AMD is also spending fairly extensively on that CCD.

 

[adroc_thurston]

Operating frequency of Venice is up a lot.
But I doubt that Olympic Ridge is up as much for a variety of reasons.

What would be the reasons?
Besides imaginary ones, obviously.

 

[gdansk]

What would be the reasons?
Besides imaginary ones, obviously.

Past performances.

 

[Fjodor2001]

No one cares about cinememe machines.

Perf/thread is what matters if you put a ceiling on max number of threads, and it'll be higher even for E-cores than Cinememe Thread spam.

 

[adroc_thurston]

Past performances.

Zen4 was 13% fmax off a single shrink to N5.

 

[adroc_thurston]

Perf/thread is what matters it you put a ceiling on max number of threads, and it'll be higher even for E-cores than Cinememe Thread spam.

Not a real metric, all that matters is socket throughput@rated power.