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

[DisEnchantment]

Speculate at will

 

[SarahKerrigan]

AT isn't a hobby???

Not one that seems to make you happy, at any rate. Doesn't make me happy either. But, old habits.

 

[igor_kavinski]

So I am not the only one who's baffled. It seems Igor has turned into a different person starting this month!😛

I'm dealing with a really bad cold at the moment.

 

[Doug S]

On single-thread? No it wouldn't. SPECint is barely sensitive to DRAM bandwidth at all. Essentially no single-threaded load that isn't a vector microbenchmark is going to saturate a modern CPU's memory bandwidth.

This is crap you just hallucinated.

SPECint is, however, sensitive to memory latency. Something you sacrifice using LPDDR5 versus DDR5. So I'd argue that if M4 used DDR5 DIMMs it might score 1-2% higher in SPECint, but I know @igor_kavinski doesn't want to hear that, because its the opposite of what he's spending so many posts trying to claim.

 

[igor_kavinski]

Not one that seems to make you happy, at any rate. Doesn't make me happy either. But, old habits.

I'm pretty happy. Just get annoyed when I see comparisons between Mx and Zen. Apple can get invited to the party when they develop their Prism emulator to run Win32 software.

 

[SarahKerrigan]

I'm pretty happy. Just get annoyed when I see comparisons between Mx and Zen. Apple can get invited to the party when they develop their Prism emulator to run Win32 software.

Win/ARM runs fine on ARM Macs in virtualization.

You seem weirdly hacked off at the Mac for being... well... the Mac. Like, it mostly runs MacOS and is going to keep mostly running MacOS for the foreseeable future.

 

[igor_kavinski]

SPECint is, however, sensitive to memory latency. Something you sacrifice using LPDDR5 versus DDR5.

Apple's huge caches are masking that latency hit. Don't think DDR5 DIMMs would help, even 12 channels of them.

 

[igor_kavinski]

Win/ARM runs fine on ARM Macs in virtualization.

At Core i5 speeds.

 

[SarahKerrigan]

At Core i5 speeds.

Win/ARM runs at M-whatever speeds. It's native code.

 

[igor_kavinski]

Win/ARM runs at M-whatever speeds. It's native code.

Link?

 

[SarahKerrigan]

Link?

Er, this isn't a controversial statement????? Why would it run at anything less? There's a little bit of overhead with virt, especially I/O, but not much.

 

[SarahKerrigan]

Link?

But since you ask so nicely...

MacBook Pro M3 Max (Parallels, ARM64, 12 cores, 25992) - Geekbench

Benchmark results for a Parallels International GmbH. Parallels ARM Virtual Machine with an Apple Silicon processor.

browser.geekbench.com

Note that the MT score here is for only 12 cores, not the full 14-16; it is unclear which ones are passed through to the guest.

And, for a comparison, the highest performing Meteor Lake ST result I could find in five minutes; I invite you to find a better one.

LENOVO 83DN - Geekbench

Benchmark results for a LENOVO 83DN with an Intel Core Ultra 9 185H processor.

browser.geekbench.com

 

[Thibsie]

I like how you answered your own question. Let me help: M4 can be pushed further. Apple chose not to, to keep it within the efficient part of the v/f curve.

Zen5 can't solve all the worlds problems too, I wish it did.

Sure, you just turn the knob and that's it. Look how it made Intel look like...

 

[Abwx]

But since you ask so nicely...

MacBook Pro M3 Max (Parallels, ARM64, 12 cores, 25992) - Geekbench

Benchmark results for a Parallels International GmbH. Parallels ARM Virtual Machine with an Apple Silicon processor.

browser.geekbench.com

Note that the MT score here is for only 12 cores, not the full 14-16; it is unclear which ones are passed through to the guest.

MT score is worthless in GB6, beside the 12 cores device tested by NBC score about 10% lower than what you posted.

Not that NBC are 100% accurate but that s a more reliable number than a GB6 submission whose conditions are unknown.

Apple M4 SoC analysis - AMD, Intel and Qualcomm currently don't stand a chance

Notebookcheck has tested the new Apple M4 processor within the iPad Pro 11 as well as the iPad Pro 13.

www.notebookcheck.net

 

[CouncilorIrissa]

MT score is worthless in GB6, beside the 12 cores device tested by NBC score about 10% lower than what you posted.

Not that NBC are 100% accurate but that s a more reliable number than a GB6 submission whose conditions are unknown.

Apple M4 SoC analysis - AMD, Intel and Qualcomm currently don't stand a chance

Notebookcheck has tested the new Apple M4 processor within the iPad Pro 11 as well as the iPad Pro 13.

www.notebookcheck.net

Yes, the M3 Pro tested by NBC scores lower than the higher-end model, the M3 Max. Big news. M3 Pro has TWICE as few performance cores.

 

[Abwx]

Yes, the M3 Pro tested by NBC scores lower than the higher-end model, the M3 Max. Big news.

Sarah Kerrigan linked a 12C GB6 score and i linked a test of a 12 cores, is there something that is wrong..?..

 

[Elfear]

Can we pretty please get back to Zen 5? I promise to burn incense in front of my wife's Macbook this evening.

 

[Nothingness]

MT score is worthless in GB6, beside the 12 cores device tested by NBC score about 10% lower than what you posted.

Not that NBC are 100% accurate but that s a more reliable number than a GB6 submission whose conditions are unknown.

Apple M4 SoC analysis - AMD, Intel and Qualcomm currently don't stand a chance

Notebookcheck has tested the new Apple M4 processor within the iPad Pro 11 as well as the iPad Pro 13.

www.notebookcheck.net

You are comparing a tablet running iOS against a Mac running Windows virtualized on MacOS. What is this supposed to show?

 

[SarahKerrigan]

MT score is worthless in GB6, beside the 12 cores device tested by NBC score about 10% lower than what you posted.

Not that NBC are 100% accurate but that s a more reliable number than a GB6 submission whose conditions are unknown.

Apple M4 SoC analysis - AMD, Intel and Qualcomm currently don't stand a chance

Notebookcheck has tested the new Apple M4 processor within the iPad Pro 11 as well as the iPad Pro 13.

www.notebookcheck.net

Congratulations! As is your tradition, you have completely missed the point!

(Igor was saying that virtualized Win/ARM has "i5 performance.")

 

[Abwx]

You are comparing a tablet running iOS against a Mac running Windows virtualized on MacOS. What is this supposed to show?

It happen that the virtualised Windows submitted at GB yield a higher score than MacOS since that s what NBC is using, so what is your point.?.

 

[CouncilorIrissa]

Congratulations! As is your tradition, you have completely missed the point!

(Igor was saying that virtualized Win/ARM has "i5 performance.")

Someone should tell people that their Windows install is likely running virtualized on top of a type-1 hypervisor. At least it does out-of-the-box IIRC.

 

[SarahKerrigan]

It happen that the virtualised Windows submitted at GB yield a higher score than MacOS since that s what NBC is using, so what is your point.?.

What specific numbers are you talking about?

 

[StefanR5R]

If only they've bothered to do GMI-Wide, then Zen5 would probably interest me a bit,

Maybe the alleged Ryzen 9 9700X sample was a cancelled GMI-wide variant of Ryzen 7 9700X. ;-)

 

[Nothingness]

It happen that the virtualised Windows yield a higher score than MacOS since that s what NBC is using, so what is your point.?.

Reread my post carefully with your finger to be sure you got all the words. See? Laptop vs tablet? Windows on MacOS vs iOS? You don't see the stupidity of the comparison?

Edit: sorry I focused on the M4 score rather than looking at the other scores. The differences between OS and version of the app might still explain the difference.

 

[SarahKerrigan]

It happen that the virtualised Windows submitted at GB yield a higher score than MacOS since that s what NBC is using, so what is your point.?.

Comrade, what are you talking about?

By your own link, M3 does ~3100 GB6 ST native. The virtualized submission above was 2697.

By your own link, M3 Pro - not Max - does 14412 with 11 cores (5 P, 6 E). The virtualized NT/ARM submission does 16165. This may have something to do with the fact that the M3 Pro 11-core SKU has only five P-cores, while the M3 Max has a minimum of ten.

Just a thought.

 

[Abwx]

Reread my post carefully with your finger to be sure you got all the words. See? Laptop vs tablet? Windows on MacOS vs iOS? You don't see the stupidity of the comparison?

Does this make a difference in GB, on average.?.

What specific numbers are you talking about?

I was talking of the MT score, to get back on topic Strix Point@30W score about 13600 pts in the less than ideal GB 6.