Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

You are hilarious.

it's trve, CAD stuff is very antiquated software to which there's no real solution for perf issues.
(but EDA stuff is worse, you should try running Vivado on poverty systems for fun)

 

[Timorous]

Things have gone too far, the AT forum hype train has reached Germany and we are good for a world tour in no time, next target is the moon.

https://twitter.com/x/status/1773772359586963909

I thought the hype was already to the moon.

 

[adroc_thurston]

I thought the hype was already to the moon.

not in mainstream perception no.
The expectation is ~z3-ish bump.
15-20% PPC with 2-3% clk bump or therein.

 

[Fjodor2001]

I thought the hype was already to the moon.

Now can someone just refer to the 3DCenter.org tweet in this thread and use that as proof of that the 40% is correct. We need to close the hype loop.

 

[Rigg]

Why would they reduce price, transistor pricing scaling is dead. Inflation globally is still high. Like what fantasy land do you al live in. Did you not just watch what happened to GPUs?

Why would amd give up margin? The answer is the people who want a 8 core zen5 buts priced to high will complain and then either buy it or buy the 6 core.

Did you not just watch what happened with gen on gen AMD CPU pricing? 5950X MSRP = $800. 7950X/X3D MSRP = $700. Zen 4 also saw street price cuts from MSRP less than 2 months after release. AMD have steadily and significantly reduced street price over time even on their current gen products. This is reality and directly contradicts pretty much everything you've just said as it relates to AMD pricing and margin. You're the one living in fantasy land.

 

[Timorous]

Now can someone just refer to the 3DCenter.org tweet in this thread and use that as proof of that the 40% is correct. We need to close the hype loop.

A hyperloop if you will.

 

[CouncilorIrissa]

Now can someone just refer to the 3DCenter.org tweet in this thread and use that as proof of that the 40% is correct. We need to close the hype loop.

That's exactly what would've happened had they not mentioned the source and the poster aka the WCCFkek approach.

 

[adroc_thurston]

AMD have steadily and significantly reduced street price over time even on their current gen products

Because comp pressure.
Which Zen5 has none of.
Not until 2026.

 

[Markfw]

Because comp pressure.
Which Zen5 has none of.
Not until 2026.

Somebody said AMD has competition now. Well, in some areas... If you declock it so it doesn't lock up.... and you have a cooling solution to cool the sun.... And you have a nuclear reactor in your back yard to power it....

(only a little sarcasm, mostly true)

 

[Doug S]

A14 big core = 2.1 mm2
A13 big core = 2.61 mm2

Please tell me where the 1.8x logic density increase is in the actual chips?

You realize they aren't the same core, right? But that's not the real problem with this irrelevant info you posted, hilariously believing it to be some sort of "gotcha".

First of all, where did you get the figures for the core sizes, and what levels of cache do they include? Even if those leave out L1 (which I doubt, almost every analyst firm includes L1 in core size) there are plenty of other structures that while they aren't "cache", will suffer the same scaling issues because they are also made from SRAM cells: register files, ROB and various other buffers, TLB, and so forth. Collectively those add up to some fairly serious size in a modern CPU - especially Apple's fairly aggressive designs - and often get bigger during generational updates as well. Either way this would inhibit scaling at anywhere near the full amount a block composed of pure logic would even if the cores were IDENTICAL in all ways, which again they are not.

Honestly, it is as though you don't understand what a CPU core is composed of and why expecting it to scale at the rate logic does is foolish.

 

[adroc_thurston]

You realize they aren't the same core, right?

Yeah, Firestorm had sizeable improvments FP-side which are pretty hungry on area.
You're wasting your time talking to him.

 

[igor_kavinski]

it's trve, CAD stuff is very antiquated software to which there's no real solution for perf issues.
(but EDA stuff is worse, you should try running Vivado on poverty systems for fun)

That's what I was thinking too. They don't have much real competition in their space so there's not really an incentive for them to optimize or rewrite their software for better performance. They know that their users have put up with their lousy software so far and will continue to do so for the foreseeable future.

 

[Timmah!]

Not even the slightest idea about niche meme applications.

Gonna remain a software problem.

So CAD is niche meme? Ok then. I guess all those engineers around the world who work with that stuff did not catch that "meme".

 

[adroc_thurston]

So CAD is niche meme?

From the perf POV? yea.

I guess all those engineers around the world who work with that stuff did not catch that "meme".

None of them have a choice of using a non- redacted piece of s/w.

No profanity in the tech forums.

CPU Mod DAPUNISHER.

 

[Abwx]

I thought the hype was already to the moon.

Yes, but we had to get back to earth to embark some carburant providers to get even further, moon is no more enough...

AMD Zen 5 CPU Core Architecture Allegedly More Than 40% Faster Than Zen 4 Cores

AMD's Zen 5 CPU core architecture might be shaping up to be a huge upgrade over the existing Zen 4 core as per a new rumor.

wccftech.com

AMD's Zen 5 core is more than 40% faster than Zen 4.....The new information comes from hardware insider, @Kepler_L2, who has a good track record when it comes to CPU architectures. According to his recent post at the Anandtech tech forum, spotted by 3DCenter,

 

[gdansk]

Look what you've done

 

[inf64]

Wow things got out of control really fast LOL.

 

[Hitman928]

Look what you've done

Live reaction cam. . .

 

[CakeMonster]

What did you expect with the current state of hardware journalism

I wonder if the usual suspects will add more rumors to get attention back to them now.

 

[Timorous]

Live reaction cam. . .

Now we just need RGT and MLID to post videos about it.

 

[Abwx]

Live reaction cam. . .

I dont like memes at all, but one cant deny that they are very creative about thoses rumours at WCCF TROLLs Inc forums...

 

[tamz_msc]

You realize they aren't the same core, right? But that's not the real problem with this irrelevant info you posted, hilariously believing it to be some sort of "gotcha".

First of all, where did you get the figures for the core sizes, and what levels of cache do they include? Even if those leave out L1 (which I doubt, almost every analyst firm includes L1 in core size) there are plenty of other structures that while they aren't "cache", will suffer the same scaling issues because they are also made from SRAM cells: register files, ROB and various other buffers, TLB, and so forth. Collectively those add up to some fairly serious size in a modern CPU - especially Apple's fairly aggressive designs - and often get bigger during generational updates as well. Either way this would inhibit scaling at anywhere near the full amount a block composed of pure logic would even if the cores were IDENTICAL in all ways, which again they are not.

Honestly, it is as though you don't understand what a CPU core is composed of and why expecting it to scale at the rate logic does is foolish.

They aren't the "same" core but whatever changes were made from A13 to A14 mainly affected fp performance, and that too in SPECfp2006 with outliers like libquantum.

On the integer side of things, you got an 18% bump in performance with a 12% increase in f_max on the big core.

The iPhone 12 & 12 Pro Review: New Design and Diminishing Returns

www.anandtech.com

A13 die block sizes are from here:

The Apple iPhone 11, 11 Pro & 11 Pro Max Review: Performance, Battery, & Camera Elevated

www.anandtech.com

A14 die block sizes are from TechInsights.

Honestly, it is as though you don't understand what a CPU core is composed of and why expecting it to scale at the rate logic does is foolish.

Tell that to the other poster who believes that TSMC advertising a 1.8x density increase from N7 to N5 means that the logic of every design scales by that factor, whether it is a big x86 core or a puny A53 or a mid-sized A78.

 

[Abwx]

Lol, once i posted the meme the relevant guy just popped up for the follow up, that s really surnatural, that s true live quantum and mystic computing combined...

 

[Joe NYC]

Did you not just watch what happened with gen on gen AMD CPU pricing? 5950X MSRP = $800. 7950X/X3D MSRP = $700. Zen 4 also saw street price cuts from MSRP less than 2 months after release. AMD have steadily and significantly reduced street price over time even on their current gen products. This is reality and directly contradicts pretty much everything you've just said as it relates to AMD pricing and margin. You're the one living in fantasy land.

Here are the current prices and how a $499 Zen 5 would fit in:

7700X - $399 (Discounted to $269 at MicroCenter)
7800X3D - $449 (Discounted to $349 at MicroCenter)
9800X - $499 (base model no V-Cache)

I don't really see the point, of AMD provoking ire of customers and reviewers by unrealistically high price, only to discount it back to more "normal price a couple of months later.

AMD would still be printing money selling the CPU at $399 to $449 MSRP and not discounting...

 

[Timorous]

AMD have never been afraid to hike prices when they have the best product. They did it in the athlon 64 FX days, they did it a bit with Zen 3 and if Zen 5 has no competition they will do it again.