Question Zen 6 Speculation Thread

[Markfw]

Here is where I would disagree. Operating Systems do not make a difference to the CPU IPC nor to its performance drastically. Macs when on Intel had the exact same IPC and performance as macOS when the CPUs were tested on Windows on the same Mac.

Likewise on Linux vs Windows. If a CPU is slower on Windows than Linux it’s because optimisation or a special flag is turned on Linux. Inherently, barring Linux distros that optimise for Zen. Operating systems do not make a difference. I mean I daily drive Windows 11 and CachyOS and there’s barely any noticeable performance difference.


No ones telling to switch to Apple. Most who mention others like RISC-V or ARM are saying it’s not the end of the road for single threaded performance for x86. After all if ISA doesn’t matter, then x86 still has major growth.

I don't know the reasons why, but th OS is significantly slower on windows than linux over many of the apps I use, mostly DC(distributed computing) applications and F@H using video.

 

[desrever]

Saying that IPC is reduced by the operating system is like saying that my car's horsepower goes down when your mom gets in. The power isn't reduced, there's just a heavier load so it goes slower.

My mom drives slower than me. My car would get slower with my mom driving. HP doesn't matter, you are comparing km/h in this case.

 

[MS_AT]

I don't know the reasons why, but th OS is significantly slower on windows than linux over many of the apps I use, mostly DC(distributed computing) applications and F@H using video.

But the OS is not making the CPU slower per se (unless the OS will disable HT like BSD for security for some CPUs, still the CPU is not slower, it has functions disabled but if you disable them in other OS then well, they will match).

Since lots of software is dynamically linked the quality and trade offs done by system libraries matter. Plus Windows has much more running in the background, you can say its tellemetry but in addition to calling home by default Windows enables now a suite of virtualization based security features which also add up when it comes to performance penalties.

Still it's not like AMD built Zen5 with a microde check if on windows, throttle😉

I think some people in this thread are confusing CPU performance with overall performance we get from the specific platform. The latter is affected by many more things than CPU alone. It's in the best interest of CPU vendor to support OS/software vendor to extract the most of the performance a cpu offer.

Since this was CPU specific thread I think we should focus on CPUs alone, otherwise we can start producing examples where library misconfigurstions can eat whole gen on gen performance advantages and well focusing on that optics it really does not matter if Zen6 will bring 10% or 5% ipc advantage.

 

[DisEnchantment]

You make a valid point. But I guess Windows can do such things to, no? I'd expect all OS to have very similar sets of mitigations, though I might be wrong.

The last Phoronix article measuring mitigation on Zen 5 showed basically no difference:

AMD Zen 5 Not Affected By Inception/SRSO, mitigations=off Yields No Benefit On Ryzen 9000 Series - Phoronix

www.phoronix.com

I noticed when we ported some CVE patches to our Zen 2 dev clusters, performance difference between the Zen 2 and Zen 4 clusters was very large than what I would expect. We rolled back the patches and it was more less what you would get with cumulative perf uplift promised by AMD.
Some of the mitigations were simply poorly done.
Obvious aggressive flushing of OoO data, caches etc would mean the processor would be stalling more.
Reducing IPC to simply the number of instructions executed would be odd considering there are lots of stalls during normal execution of code which is why optimization guide exist in the first place. There are hazards that needs to be handled, optimum data structure, etc..
The OS, Hypervisor, User space libs etc in your SW env matters

 

[CouncilorIrissa]

I don't think Apple M runs Windows with very good IPC, right?. I don't give two sh*ts about Apple and their confounding and restrictive OS. It's great for some people, but not me. I like to be able to "get in the box" and run literally millions of apps from yesterday and 50 years ago. Windows runs everything.. from apps designed by hundreds of highly qualified engineers to smart kids in their basements. Apple is East Berlin before the wall came down. Everything is controlled and set up for Apple. x86 has to do it all. Yes, freedom is harder to lock down for optimizing. Apple? No thanks.

In addition, this is a Zen 6 thread.

Their cores perform exceptionally well when benchmarked on Linux.


Current x86 cores are just bad client cores, plain and simple. In what world it is acceptable for a laptop core to be gapped by a phone (the iPhone, to be precise) in running browsers.

 

[Joe NYC]

What do you mean client workloads don't matter? Client was over 25% percent of AMD's 2024 revenue.

And chances are, AMD client CPU revenue will surpass datacenter CPU revenue again this year.

 

[DisEnchantment]

And chances are, AMD client CPU revenue will surpass datacenter CPU revenue again this year.

And from the patent trail from last quarter you'd imagine AMD is a packaging specialist than CPU designer. Not good. Typically patent trail is good indicator of active research in novel ideas but not since last two quarters unfortunately. At least not much applications in core design.

 

[Hulk]

If they both fail that hard they simply will not be competitive for consumer workloads.

5% IPC ST increase and better MT will be better than what is available now and people seem to be getting along with their consumer workloads just fine. I don't understand your statement.

The reality is there is no other x86 manufacturer to turn to.

 

[CouncilorIrissa]

5% IPC ST increase and better MT will be better than what is available now and people seem to be getting along with their consumer workloads just fine. I don't understand your statement.

The reality is there is no other x86 manufacturer to turn to.

People, especially younger ones, already tend to use their phones as their main computing device. If phones start beating laptops in consumer workloads, then the whole segment will fade into irrelevance (DT already did).

There will be no need for x86.

 

[Hulk]

Here is where I would disagree. Operating Systems do not make a difference to the CPU IPC nor to its performance drastically. Macs when on Intel had the exact same IPC and performance as macOS when the CPUs were tested on Windows on the same Mac.

Likewise on Linux vs Windows. If a CPU is slower on Windows than Linux it’s because optimisation or a special flag is turned on Linux. Inherently, barring Linux distros that optimise for Zen. Operating systems do not make a difference. I mean I daily drive Windows 11 and CachyOS and there’s barely any noticeable performance difference.


No ones telling to switch to Apple. Most who mention others like RISC-V or ARM are saying it’s not the end of the road for single threaded performance for x86. After all if ISA doesn’t matter, then x86 still has major growth.

Good point about when Apple was on x86 and they didn't seem to have an OS-App tie performance advantage. Hadn't thought of that.

But I would counter with the fact that the lastest OS and the M series were designed with one another in mind and that is a possibly insurmountable advantage versus x86.

Also, as always Apple doesn't have to service an enormous legacy code base, which is why they have been able to switch OS like 5 times in the last 35 years.

But your points stands. Apple OS and M do set a bar for IPC. No doubt about that.

It's a good point against my argument that all of the low hanging fruit is gone!

I can admit when when someone "poke" hole in my case! Yes, pun intended.

 

[Hulk]

Their cores perform exceptionally well when benchmarked on Linux.
View attachment 119380

Current x86 cores are just bad client cores, plain and simple. In what world it is acceptable for a laptop core to be gapped by a phone (the iPhone, to be precise) in running browsers.

Yes as I replied to poke01 that is a good point against my argument. I must admit.

Maybe there is hope!

 

[poke01]

There will be no need for x86.

Servers and legacy will need x86 and gaming (PS5, Xbox)

Plus on desktop, x86 is the only platform where it’s upgradable.

 

[poke01]

As for Zen6, Nanoflex on N2 will remove the frequency stagnation. Fun times ahead.

 

[CouncilorIrissa]

Servers and legacy will need x86 and gaming (PS5, Xbox)

Plus on desktop, x86 is the only platform where it’s upgradable.

Mobile gaming has already surpassed PC/console gaming in revenue. And hardcore gamers are much more loyal to Huang's gigarays than they are to CPU manufacturers, so all it takes for them to be put into a very precarious position is for Huang to come up with a decent ARM SoC (which they're working on with Mediatek).

 

[adroc_thurston]

Mobile gaming has already surpassed PC/console gaming in revenue

Casinos are not video games.

 

[gdansk]

5% IPC ST increase and better MT will be better than what is available now and people seem to be getting along with their consumer workloads just fine. I don't understand your statement.

The reality is there is no other x86 manufacturer to turn to.

They are not getting along just fine. Outside of the server space sales are down.

Their competition delivers more than that every year. There is little chance any enthusiast will be staying if they deliver only 5% IPC. They have to deliver better or they will vacate the consumer space (and in a way, voluntarily).

 

[adroc_thurston]

Outside of the server space sales are down.

?

Their competition delivers every year

They'll run out of shrinks like everyone else.
Can ship a 5GHz core once!

 

[gdansk]

They'll run out of shrinks like everyone else.
Can ship a 5GHz core once!

And yet they're pulling away since x86 peaked with Zen 3 and hasn't kept up since.

 

[adroc_thurston]

And yet they're pulling away since x86 peaked with Zen 3 and hasn't kept up since.

Zen4 was a major fmax bump too!
5.05-ish to 5.7 is 12% fmax.
It's mobile where AMD hasn't seen freq bumps since RMB.

 

[Markfw]

And yet they're pulling away since x86 peaked with Zen 3 and hasn't kept up since.

and who is this magical competitor that is pulling away. You never answered my last post either. Are you just trolling ?

 

[gdansk]

and who is this magical competitor that is pulling away. You never answered my last post either. Are you just trolling ?

You know who it is. Stop pretending that just because they're a lifestyle company they can't make good CPUs.

And go look at my original post. There is only one CPU series with a SPECint 1T score high enough and clock rate low enough to be nearly 50% ahead per GHz.

 

[Kepler_L2]

IPC is not a property related in any way to the OS!

In theory no, but Zen is like 10-20% faster on Linux.

 

[Kepler_L2]

People, especially younger ones, already tend to use their phones as their main computing device. If phones start beating laptops in consumer workloads, then the whole segment will fade into irrelevance (DT already did).

There will be no need for x86.

"PC is dying" is a 20-year old trope and never been true. In fact latest research shows the largest growth market for things like gaming is PC.

 

[gdansk]

"PC is dying" is a 20-year old trope and never been true. In fact latest research shows the largest growth market for things like gaming is PC.

PC has, despite all the hate over many years, continually increased performance. And so it had the highest available performance for over 20 years. Until it was finally passed in 2024. If - like Hulk is saying - 5% after 2 years is the best they can ship do you think it can still retain its relevance?

At least AMD's slide for Zen 6 was more optimistic than that.

 

[poke01]

That slide could be red herring, there is no way that AMD only manages only 5-10% IPC in 2 years on N2 and if they do it’s because of Zen architecture needing a revamp and not cause x86 is dead.