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

[DisEnchantment]

Speculate at will

 

[Glo.]

Cue the OEM desktop with weaker dGPU and single channel memory. Mandatory AI label though, those TOPS ain't gonna sell themselves!

Bam! 546 million laptops sold in this config!

Done!

All thanks to AI PC branding!

 

[Thibsie]

Let's hope so.

 

[Nothingness]

This is not from Andrei F. Its from some guy who can't quote. A terrible read.

And it predates the release of CB24 which was better tuned for Apple Silicon than previous releases.

 

[soresu]

Cinebench uses Intel Embree Engine underneath. What do you think it's optimized for?

Read this on why Cinebench is a poor standard CPU benchmark from Ex-Anandtech Andrei F:

https://www.reddit.com/r/hardware/comments/pitid6

The reason why Cinebench is so popular is because AMD used it in its main marketing because it was the main benchmark earlier Zen chips excelled in. Cinebench does not correlate with most consumer workloads.

I smell denial.

It doesn't matter what benchmark you choose to make your favourite.

No single benchmark is fully representative of performance.

That's why any good reviewer uses multiple benchmarks.

Web browsers alone have several.

 

[FlameTail]

This discussion.

 

[adroc_thurston]

A process with 60% yields is broken.

But it's not 60% yield?

Cinebench uses Intel Embree Engine underneath. What do you think it's optimized for?

Not the r24, christ.

Read this on why Cinebench is a poor standard CPU benchmark from Ex-Anandtech Andrei F:

https://www.reddit.com/r/hardware/comments/pitid6

"please listen to another ARM shill. onegai."
not sending their best eh

The reason why Cinebench is so popular is because AMD used it in its main marketing because it was the main benchmark earlier Zen chips excelled in. Cinebench does not correlate with most consumer workloads.

It's popular because it's an easy-to-run throughput workload.

 

[moinmoin]

The reason why Cinebench is so popular is because AMD used it in its main marketing because it was the main benchmark earlier Zen chips excelled in. Cinebench does not correlate with most consumer workloads.

AMD used it for the Zen 1 launch since before that point Cinebench was a typical benchmark to test Intel chips.
Google search result from before December 2016.

 

[Doug S]

• As for yield issue, CEO of TSMC has addressed in Q3 2023's financial seminar:- "N3 is already in volume production with good yield and we are seeing a strong ramp in the second half of this year, supported by both HPC and smartphone applications"

You want to believe what the rumored said or CEO of TSMC CEO said?

Unless the claims of Apple paying per known good die are untrue, it is pretty clear N3B is much worse than any TSMC process since Apple became a customer. Of course the CEO is going to paint a rosy picture.

We'll see how quickly Apple ditches anything N3B related and that will tell. The images of M3 Max that appear to show it without the interdie connection to make an Ultra are another potential clue. If only M3 Max loses that and M4 gains an 'Ultra' version, it will be pretty clear why that is.

 

[Tigerick]

Unless the claims of Apple paying per known good die are untrue, it is pretty clear N3B is much worse than any TSMC process since Apple became a customer.

• As for yield issue, CEO of TSMC has addressed in Q3 2023's financial seminar:- "N3 is already in volume production with good yield and we are seeing a strong ramp in the second half of this year, supported by both HPC and smartphone applications"

You want to believe what the rumored said or CEO of TSMC CEO said?

 

[soresu]

Cinebench uses Intel Embree Engine underneath. What do you think it's optimized for?

Embree is important for sure, but it is NOT the renderer itself.

Also they may have originally used Embree wholesale for the sake of simplifying the initial codebase development - but that doesn't tell us the absolute internal structure of the Cinebench renderer today

They could have long since transitioned to mostly custom RT kernels, the same way that Valve's Source engine still had bits and pieces of GoldSrc wedged in there, or Freespace Open still has truly ancient Volition codebase bones sticking out of its shiny PBR rendering space engine shell.

As long as even part of the codebase still has Embree code in there they have to declare it.

 

[soresu]

View attachment 93803
This discussion.

That is somehow both hilarious and extremely disturbing.....

 

[Thibsie]

• As for yield issue, CEO of TSMC has addressed in Q3 2023's financial seminar:- "N3 is already in volume production with good yield and we are seeing a strong ramp in the second half of this year, supported by both HPC and smartphone applications"

You want to believe what the rumored said or CEO of TSMC CEO said?

Mmm how is N3E progressing?
They talk about N3, not N3B or N3E.
Could he talk about N3 in general or this has to be N3B ?

 

[adroc_thurston]

They talk about N3, not N3B or N3E.

N3 means N3b.
N3e is a different node altogether, they call it 2nd gen N3 in their official PR or something like that.

 

[CakeMonster]

Any news on the integrated graphics compared to Z4? I don't really care about performance, since everyone who has been following this market for a number of years no longer believes it will ever get 'good enough' for their fantasies of an IGP, but I am curious about improved features for video etc.

 

[adroc_thurston]

but I am curious about improved features for video etc.

It uses pretty much the same VCN.
Nothing changed.

 

[soresu]

Any news on the integrated graphics compared to Z4? I don't really care about performance, since everyone who has been following this market for a number of years no longer believes it will ever get 'good enough' for their fantasies of an IGP, but I am curious about improved features for video etc.

Phoenix Point already has everything but VVC and EVC encoder and decoder of the modern video codecs.

VCN5 is apparently coming for RDNA4 - exactly what is in there is anyones guess.

 

[Thunder 57]

View attachment 93803
This discussion.

At least no one is saying "Userbenchmark best!".

 

[mikegg]

It doesn't matter what benchmark you choose to make your favourite.

No single benchmark is fully representative of performance.

That's why any good reviewer uses multiple benchmarks.

Web browsers alone have several.

Agreed. But why are we pretending like Cinebench isn't the #1 CPU benchmark for AMD and Intel? And it's what notebookcheck uses to base their perf/watt as well as general perf. That's what sparked this discussion.

 

[mikegg]

It's popular because it's an easy-to-run throughput workload.

And absolutely meaningless when it comes to measuring CPU performance.

 

[mikegg]

That is not correct. 1T results in Cinebench do not correlate with SPEC 1 rate when it comes to Raptor Lake and Zen 4 (which are basically neck and neck in Spec but Raptor Lake leads in Cinebench). I'm referring to the performance per clock.

Cinebench doesn't correlate with most applications. It shouldn't even be used as a CPU benchmark, let alone the main one right now like most Youtubers, Ananadtech AMD users, and notebookcheck.

 

[FlameTail]

We should all strive to be the top guy.

 

[dr1337]

And absolutely meaningless when it comes to measuring CPU performance.

So what benchmarks are better? Because as far as I can tell Cinebench is 2nd only to the Blender benchmark in terms of real world performance impact.

Microbenchmarking isn't without its merits but I have pretty much never found geekbench numbers to correlate much to anything I do on a computer/device. My iPhone gets higher geekbench scores than an old haswell rig, but when trying to do things like encode a video (yeah I tried to CPU encode on a phone, fight me), it's obvious the iPhone SOC was not even remotely close to the PC in terms of performance.

 

[StefanR5R]

Cinebench doesn't correlate with most applications. It shouldn't even be used as a CPU benchmark,

Most Benchmarks don't correlate with most applications. If you don't like that folks present Cinebench measurements, than you may also don't like that SPECrate®2017
Floating Point users don't shy away from CPU image rendering benchmarks either. Edit: OMG, Geekbench 6 includes an Intel Embree based ray tracer too. :-O

Can't you guys (including the meme poster) give it a rest already and simply take individual benchmarks for what they are? And how about we all went back to the thread's topic for a change?

 

[Nothingness]

So what benchmarks are better? Because as far as I can tell Cinebench is 2nd only to the Blender benchmark in terms of real world performance impact.

Both these benchmarks are good for users that care about software rendering. That probably is less than 1% of computer users. That looks like a poor choice to assess "real world performance". You need several benchmarks for that.

 

[DisEnchantment]

no?
They've been hiring compiler and linux engineers for a reason.

The Zen 4 enablement is not indicative since it wasn't really different form the rest of Family 19h from the machine PoV. OTOH the Zen 3 situation was similar to the Family 1Ah/Zen 5 one. Yet the Zen 3 GCC patches were published early Dec 2020 - a month after the launch.

Seeing both binutils and GCC posted like 2Qs before launch is certainly an improvement. However, when you count the real "end user availability" there is a room for improvement... Post patches for a toolchain component -> wait till merged -> wait till released -> wait till a mainstream distro upgrades (-> wait till somebody actually recompiles packages using it).

For AMD, they don't usually upstream compiler patches unless the CPU is launched.
Forget the patches, you wont even see the manual or the optimization guide.

Intel usually have manuals for future generations years before launch.

So it is indeed unusual for AMD to release patches before CPU availability.

There are some exceptions though when they tried upstreaming to kernel for bring up, they got blocked because maintainers are not ready to accept changes without a manual and AMD was forced to add some vague sections in the reference manual.