Question Zen 6 Speculation Thread

[soresu]

Also brand loyalty does not exist in CPU spaces.

It sooooo does.

I've seen it from at least 2 different generations of fairly tech savvy people separated by about 40 years.

I have it myself having not bought a single Intel CPU since my 3 Ghz P4/Prescott in.... I wanna say 2004?

(I feel oooolllldd, get me my walking cane 👴🩼)

Though it's for a different reason than the typical "I like this brand, so I buy this brand" mentality.

It's more like "I don't like this brand, so I buy the only other brand" kind of mentality.

Same reason I buy AMD instead of nVidia in the GPU space since 2009.

 

[adroc_thurston]

It sooooo does.

MSS data proves otherwise.

 

[Thunder 57]

?

Your claim that nobody uses iGPU's.

 

[adroc_thurston]

Your claim that nobody uses iGPU's.

Yeah, in market terms, iGFX doesn't matter at all.
AMD spent the better part of the previous decade shipping far better GPUs than Intel and it helped not even an iota.

 

[HurleyBird]

That is true but for most productivity tasks there is little difference between a 9700X and 9800X3D. I have seen some on more thorough sites like Phoronix but those tend to be things your average user doesn't care much about.

Single digit difference most of the time, but for the first time the difference tends to be in the 3D variant's favor. That means that only having only a single CCD with V-cache on the 9950X3D is not longer about getting "the best of both worlds," but AMD pinching pennies.

But exceptions exist, and making those exceptions awkward is.... awkward.

And 8-core parts aren't bottlenecked by bandwidth. Large caches also relieve bandwidth pressure.

If anything we should expect V-cache to provide an greater benefit for Zen 6 given further iterative evolution and 50% higher capacity on the V-cache side vs. 50% more cores, ~15% higher clocks, and a bunch more stuff potentially competing for bandwidth on the IOD.

Yes the memory controller is also improving, but not proportionately, and bleeding edge fast memory is expensive.

 

[poke01]

Yeah, in market terms, iGFX doesn't matter at all.
AMD spent the better part of the previous decade shipping far better GPUs than Intel and it helped not even an iota.

They don’t?? IGPU is reason why AMD has Sony and MS contracts.

And they are the reason why I would recommend Panther lake over Strix point refresh/Qualcomm.

iGPUs matter now. They didn’t a decade ago cause no one cared then.

 

[Thunder 57]

Yeah, in market terms, iGFX doesn't matter at all.
AMD spent the better part of the previous decade shipping far better GPUs than Intel and it helped not even an iota.

It was always going to be hard for them to get into laptops and prebuilts without a proven track record. Same with server. FWIW the company I started at in 2020 is massive and was already issuing Elitebooks with 3500U's. That was before they started to take off with Renior. Also companies have contracts they can't just switch at a moments notice.

 

[HurleyBird]

Yeah, in market terms, iGFX doesn't matter at all.
AMD spent the better part of the previous decade shipping far better GPUs than Intel and it helped not even an iota.

Had it been Intel shipping far better iGPUs AMD might not be around today.

But yes, after a certain point it doesn't matter how many times you lap the competitor's iGPU when your CPU is otherwise uncompetitive.

iGPUs mostly only matter today to the extent they can accelerate local AI models, which at the moment rounds to 0 for economy (non-ultra/max/halo) iGPUs.

 

[adroc_thurston]

IGPU is reason why AMD has Sony and MS contracts.

Wrong, they won 'em because their big boy GFX is good.
PS4 was like a castrated 7850 for the most part.

iGPUs matter now.

nnnnope.

It was always going to be har for them to get into laptops and prebuilts without a proven track record. Same with server

They had a proven track record, AMD is (was) a major x86 CPU vendor even then.
It's just that AMD sucked at CPU and battery life, and turns out you need good CPU and BL to make inroads into laptop.

Had it been Intel shipping far better iGPUs AMD might not be around today.

Wrong, AMD was competing on price, not iGFX grunt. Since the latter just didn't matter.
Kaveri was a GPU powerhouse for 2014, with proper GCN2 implementation on-board and it mattered not even an iota versus Haswell.

iGPUs mostly only matter today to the extent they can accelerate local AI models

Congrats on picking a usecase less relevant than video games to your average laptop consumer.

 

[LightningZ71]

It's the truth, DIY CPU marketshare did rightful swings between either vendor depending on release quality.

DIY market share? Something that you have stated is such a miniscule portion of the market that it isn't even worth tallying? DIY, the denizens of which hang out in brand specific social media groups where they groupthink their way through coping over their Brand's problems by picking apart minute corner cases of the competition that maybe 12 people on the face of the planet might notice as justification for their continuing to purchase their favorite brand's fecal matter at marked up prices?

Are there ones that don't care about brand? Certainly! But there is a notable chunk of the market that will never purchase the competition's products for various reasons, true or imagined.

And that completely ignores the massive corporate market that buys anything Intel no matter what.

 

[adroc_thurston]

DIY market share? Something that you have stated is such a miniscule portion of the market that it isn't even worth tallying? DIY, the denizens of which hang out in brand specific social media groups where they groupthink their way through coping over their Brand's problems by picking apart minute corner cases of the competition that maybe 12 people on the face of the planet might notice as justification for their continuing to purchase their favorite brand's fecal matter at marked up prices?

yeah none of that is a thing in CPU lands.

But there is a notable chunk of the market that will never purchase the competition's products for various reasons, true or imagined.

Nope.

And that completely ignores the massive corporate market that buys anything Intel no matter what.

Very untrue, AMD is finally getting their commercial push proper.

 

[LightningZ71]

In a tiny irrelevant market.

One that's growing every quarter in the last 2 years...

idk chief, 8CU is smaller than even 12CU in Phoenix.
We're so back!

Weren't you the one telling me that the 8cu iGPU in Kraken was every bit as good as the 12cu unit in Phoenix/HPT? Benchmarks show that it's a statistical dead heat. Unless Medusa is using the exact same 8cu's of RDNA3.5 on the same exact process node N4P, it's going to be a higher performance iGPU than the 12cu of Phoenix/HPT. AMD, the same ones that walked back Vega11 from the Zen1 and 1+ APUs to Renoir's 8CUs and still out performed them? It's still going to be an improvement and will likely close in on STP's iGPU because it's so bandwidth starved without MALL.

Calling that iGPU is very dishonest.

It's an integrated GPU... What more do you want?

Smaller than everything shipped since 2022?
8CUs, completely and utterly castrated versus LPDDR speeds they'll be shipping for MDS1.
Too bad!

Let's wait for benchmarks there. It remains to be seen if it's going to be slower than anything but Strix anything in any benchmark.

You don't need more than 4CUs for that anyway.

Now that, we can agree on.

 

[adroc_thurston]

One that's growing every quarter in the last 2 years...

Still tiny.

Weren't you the one telling me that the 8cu iGPU in Kraken was every bit as good as the 12cu unit in Phoenix/HPT?

Yeah, only that KRK is a mainstream part.
Now, with MDS1, you don't have gfx bump going above mainstream.

AMD, the same ones that walked back Vega11 from the Zen1 and 1+ APUs to Renoir's 8CUs and still out performed them?

t'was a way, way bigger shrink there.

It's an integrated GPU... What more do you want?

it's a 7850 with two Jaguar clusters taped to the side.

Let's wait for benchmarks there. It remains to be seen if it's going to be slower than anything but Strix anything in any benchmark.

you're getting less stuff. Are you happy you're getting less stuff?

 

[Thunder 57]

They had a proven track record, AMD is (was) a major x86 CPU vendor even then.
It's just that AMD sucked at CPU and battery life, and turns out you need good CPU and BL to make inroads into laptop.

How is having a bad CPU and poor battery life a proven track record? During the Construction Core days AMD was limited to budget models with crappy screens and such. I doubt any busniess oriented models ever existed. They may have during the K8 days but wasn't looking at laptops back then. When Zen came around they got good CPU's, iGPU's, better built laptops, and I would imagine reliability and support (RAS). And they have kept iterating on that. That is what I meant by "proven track record".

Kind how Naples saw poor adoption. But it lead to Rome with more and more adoption along the way.

 

[adroc_thurston]

How is having a bad CPU and poor battery life a proven track record?

That's an SOC design characteristic; has nothing to do with their track record as x86 vendor in general.

During the Construction Core days AMD was limited to budget models with crappy screens and such

Yeah, lesson simple: don't make suckass CPUs.

When Zen came around they got good CPU's, iGPU's, better built laptops, and I would imagine reliability and support (RAS)

the IGP has been 'good' since Llano, but again, that's an utterly irrelevant factor.

That is what I meant by "proven track record".

Oh no what you mean is AMD stopped sucking at laptop SOCs.
Which is fair.

Kind how Naples saw poor adoption

Way way way way different thing since AMD exited the server market for a brief while altogether.
And generally speaking the comp situation in server has been worse ever since Nehalem launched.

 

[adroc_thurston]

-Redacted-

That's not a representation of DIY MSS swings.

The Intel brand STILL holds major sway outside of DIY circles.

Kind of.

 

[LightningZ71]

Yeah, only that KRK is a mainstream part.
Now, with MDS1, you don't have gfx bump going above mainstream.

AMD rightly knows that it would be pointless without a hunk of MALL cache, which would be quite pricey for a mainstream part. If you want them to pointlessly waste silicon, that's a YOU issue. They pivoted to a big NPU. I don't see the value there, but that's what we get.

t'was a way, way bigger shrink there.

You still got "less stuff" there and we're happy to get it!

you're getting less stuff. Are you happy you're getting less stuff?

If it means that the part is more affordable and still is faster than it's predecessor when benchmarked, then I don't particularly care how they got there.

 

[Kepler_L2]

If it means that the part is more affordable and still is faster than it's predecessor when benchmarked, then I don't particularly care how they got there.

There is no way 8 CU RDNA 3.5+ beats 16 CU RDNA 3.5

 

[adroc_thurston]

AMD rightly knows that it would be pointless without a hunk of MALL cache

Higher LPDDR speeds exist (just in case).

You still got "less stuff" there and we're happy to get it!

Bigger shrink than going to N3p. too bad!

If it means that the part is more affordable and still is faster than it's predecessor when benchmarked

It's gonna be slower and less affordable.
You're paying for general CPU perf and BL there.
Not the graphics. Graphics does not matter.

 

[Joe NYC]

Man, I made a career out of trolling Intel dolts and tards on WCCFTECH who believed anything was possible with "muh tuning". When the Arrow Lake reviews hit it was like a bug bomb was dropped on the place, about 75% of the usual suspects/cockroaches vanished in a couple weeks, and even many of the venerable 'tards of the round table disappeared, never to be seen or heard from again. That was almost a year ago. The Intel brand STILL holds major sway outside of DIY circles.

View attachment 126799

Excellent observations. Shows there is a lot of brand loyalty, enduring instability of their (loony) tuned systems, up to a point of ridicule and surrender (with Arrow Lake).

In the meantime, there is also a lot of deferrals of purchases, until their favorite vendor releases a CPU that is not laughable. So, you will see people hanging onto their Alder Lake, Raptor Lake as long as they can, arguing that CPU performance does not matter, trying to hold out for Nova Lake.

Which, means, there will be a lot of pent-up demand when Nova Lake is released, and margin of Zen 6 vs. Nova Lake will matter. If it takes 2-hi V-Cache, AMD will release it to hold their current margin.

Lisa will have complete victory at hand, but wait, @adroc_thurston says Lisa will rather nickel and dime than go for full victory.

 

[LightningZ71]

The 16CU STP iGPU is held back by both a trailing foundry node (N4P) and by limited memory bandwidth. The next gen (not refresh) part will be on an N3 family node with the potential to also benefit from fin-flex for notably higher clocks while also being fed by incrementally or even notably faster memory. There will be cases where it bests STP, especially in bandwidth bound scenarios. There will also be cases where STP bests it when raw compute is the limiting factor. VEGA8 did this in Renoir vs. Vega 11 in earlier APUs.

Just to emphasize the point, Kraken Point already benchmarks neck and neck with 12cu Hawk Point processors, which Strix Point doesn't exactly run away from and benefits more from sustained higher clocks than anything else.

 

[adroc_thurston]

Excellent observations. Shows there is a lot of brand loyalty, enduring instability of their (loony) tuned systems, up to a point of ridicule and surrender (with Arrow Lake).

Mentally unstable forum dwellers have no relation to gamers™ at large.

says Lisa will rather nickel and dime than go for full victory.

I ain't saying it, that's what they do.
And they've been doing it rather well, and for ages.

The next gen (not refresh) part will be on an N3 family node with the potential to also benefit from fin-flex for notably higher clocks while also being fed by incrementally or even notably faster memory

You're just not getting the same perf out of 8CUs clocked generously 15% higher.
And no, no one's juicing 3-2 into a GPU.

It'll be a bump for the mainstream (albeit very tiny) and a downgrade for the premium segments.
But you'll get better BL and a better CPU in return (aka Things That Matter). Case closed.

 

[poke01]

there is loyalty in every part of every industry, it takes long repeated stuff ups to break a 30 year old pattern.

So far since Skylake Intel f upped a lot.

 

[adroc_thurston]

So far since Skylake Intel f upped a lot

Coffee/Coffee-R/Alder were good desktop releases.
Raptor was also good but then boom.

 

[Shmee]

Hello everyone, this is a reminder to stay on topic, and to be civil. Trolling will not be tolerated. -AT Moderator Shmee