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

[DisEnchantment]

Speculate at will

 

[Thunder 57]

Nothing really except we never see it released. It is a stepping/revision code. Usually silicon has gone through at least a couple of revisions before release.

 

[gdansk]

And what's the significance of A0 silicon?

Usually we don't get the first stepping

 

[Fjodor2001]

Finally the leaks we’ve been waiting for ages for start coming. Here’s another one:

AMD Ryzen AI 9 HX 370 APU ES Benchmarks Leak: Up To 25% Faster Multi & 17% Faster Single-Thread Performance In CPU-Z​

AMD Ryzen AI 9 HX 370 APU ES Benchmarks Leak: Up To 25% Faster Multi & 17% Faster Single-Thread Performance In CPU-Z

Benchmarks of AMD's Ryzen AI 9 HX 370 APU in ES state have leaked out in CPU-z, showcasing up to 25% faster performance over its predecessor.

wccftech.com

 

[Joe NYC]

Finally the leaks we’ve been waiting for ages for start coming. Here’s another one:

Still says "Engineering Sample"

 

[CouncilorIrissa]

Finally the leaks we’ve been waiting for ages for start coming. Here’s another one:

AMD Ryzen AI 9 HX 370 APU ES Benchmarks Leak: Up To 25% Faster Multi & 17% Faster Single-Thread Performance In CPU-Z​

AMD Ryzen AI 9 HX 370 APU ES Benchmarks Leak: Up To 25% Faster Multi & 17% Faster Single-Thread Performance In CPU-Z

Benchmarks of AMD's Ryzen AI 9 HX 370 APU in ES state have leaked out in CPU-z, showcasing up to 25% faster performance over its predecessor.

wccftech.com

View attachment 100821
View attachment 100822
View attachment 100823

It was discussed 4 pages ago. The most annoying part is that we don't know what clocks the ST test was run at.

 

[Fjodor2001]

Still says "Engineering Sample"

Yupp, but still it’s something more than just speculation, which we were stuck with previously. 🙂

 

[Kepler_L2]

Would this generation be so special? AMD decide to release high performance DT chips GraniteRidge A0 silicon to consumers.

https://twitter.com/x/status/1799428696694268350

View attachment 100817

That's super interesting if true. A0 was ready since early 2023, so Zen5 could have launched a long time ago. Also B0/C0 CCDs exist according to certain open source AMD stuff.

 

[inf64]

It was discussed 4 pages ago. The most annoying part is that we don't know what clocks the ST test was run at.

I think someone said it peaked at around 4.5Ghz or so.

 

[Markfw]

Still says "Engineering Sample"

And adroc was talking about 9950x single thread spec, so he still has a chance !

 

[CouncilorIrissa]

I think someone said it peaked at around 4.5Ghz or so.

4.43 if you look at the video, however the author looks away in the middle of the ST test, so we only get to see clocks at the end of the test.

 

[tsamolotoff]

4.43 if you look at the video, however the author looks away in the middle of the ST test, so we only get to see clocks at the end of the test.

It does not change (the clock), also it's the clock of the first core (and maybe 2nd or 3rd), CPU-z ignores CPPC and prevents you from changing that (unless you use special tools like BenchMate etc and launch it with predefined affinity)

 

[HurleyBird]

Not bad considering the benchmark doesn't favor Ryzens.

Don't get this talking point. At a thousand foot level, if you would expect to find large gains anywhere after making several major architectural changes, it would be in tasks that you previously performed poorly in.

 

[igor_kavinski]

it would be in tasks that you previously performed poorly in.

True. But not even clear what useful task CPU-Z is performing that normal users do. I just use it to see if overclocking the CPU is netting me any gains. It's a quick benchmark to ascertain that. I'm sure AMD doesn't give an ass whether they move the needle in CPU-Z.

 

[inf64]

True. But not even clear what useful task CPU-Z is performing that normal users do. I just use it to see if overclocking the CPU is netting me any gains. It's a quick benchmark to ascertain that. I'm sure AMD doesn't give an ass whether they move the needle in CPU-Z.

Well Zen 4 achieved an "amazing" 1% improvement in CPUz 1T so the results for Zen 5 are astonishingly good, looking from Zen 4 perspective.

 

[Abwx]

Don't get this talking point. At a thousand foot level, if you would expect to find large gains anywhere after making several major architectural changes, it would be in tasks that you previously performed poorly in.

You mean in tasks where you previously performed so well that CPU Z editor "updated" the bench to get Intel at the first place again under the pretense that Zen 1 was taking advantage of an alleged bug to score better..?..

Same with WebXprt wich is an Intel bench now edited by the infamous Principle Technologies, AMD was ahead by 10% in WebXprt 3, then come Webxprt 4 that nullified the advantage and put Intel on par.

We can also add the transition from CB R15 to R20/R23, at this rate while AMD is improving real IPC Intel get the benches to be "improved" on their behalf.

 

[Hail The Brain Slug]

Referencing any CPU-z 1T numbers for Ryzen is like using a random number generator to get benchmark results due to locking to core 0 and ignoring preferred cores. Why are we even still talking about CPU-z figures when we know they are worse than useless?

 

[yuri69]

Very Premium Thing

In the x86 notebook market this really doesn't jive with the missing GeForce RTX sticker...

 

[adroc_thurston]

In the x86 notebook market this really doesn't jive with the missing GeForce RTX sticker...

Welcome to the new world.

 

[poke01]

Sandbagged Zen 4 prior to release:

View attachment 100805

View attachment 100807


Real Zen 4 (released):

View attachment 100806

+14% headroom in clock speed (5Ghz+ versus 5.7Ghz on launch)

~8% IPC (early reveal) vs 13% on launch.

>15% ST uplift turned out to be ~29% ST uplift.

You're welcome.

The launch for Zen 5 is next month. Strix Point is launching next month. This is quite a different launch to Zen 4.

The clocks are set in place. They didn’t reveal SKUs or definite clocks last time at the Zen 4 computex.

 

[invisible_city]

So sandbagging is not real in Zen 4 case because it was 4 months before the release? LMAO

It’s not real because what they announced and delivered were within margin of error.

 

[Thunder 57]

It’s not real because what they announced and delivered were within margin of error.

If we use Intel's latest definition of margin of error, sure. +/- 10% gives plenty of wiggle room.

 

[invisible_city]

The launch for Zen 5 is next month. Strix Point is launching next month. This is quite a different launch to Zen 4.

The clocks are set in place. They didn’t reveal SKUs or definite clocks last at the Zen 4 computex.

Hope everlasting. We’re going to be hearing about secret sauce and how Zen 5 is really 50%+ because on some specific AVX512 workload it over delivers, for months to years. We’ll be hearing about sandbagging in nursing homes worldwide in 50 years. “Back in my day, AMD sandbagged so hard….it was glorious”. Ok grandpa.

 

[invisible_city]

If we use Intel's latest definition of margin of error, sure. +/- 10% gives plenty of wiggle room.

What does Intel have to do with AMD slides?

 

[poke01]

if anyone is truly sandbagging it would be Intel this time. Their CPUs don’t launch till September with Lunar and October/November for Arrow. So it’s “okay” for Intel to use the margin of error.

I do think Zen 5 will be great architecture relative to Zen 4 but it’s not some mythical core that will obliterate every other CPU microarchitecture for the next couple of years as some leakers touted it be. In that regard they already got proven wrong.

I’m sure we will see some insane gains in AVX-512 but the actual core performance will have to revealed by SPEC.

 

[branch_suggestion]

The launch for Zen 5 is next month. Strix Point is launching next month. This is quite a different launch to Zen 4.

The clocks are set in place. They didn’t reveal SKUs or definite clocks last time at the Zen 4 computex.

No price, no exact release date.
Not a complete launch.