Question Zen 6 Speculation Thread

[511]

but i want a 7GHz core at 35 watts

I want 6.9Ghz core at 6.9 Watts

 

[poke01]

I want 6.9Ghz core at 6.9 Watts

By the end of this decade we need another huge IPC increase

 

[Doug S]

? Apple got a 10% frequency increase from N3B

You're assuming they got that all from process, and none from design or from deciding to use more power in single core loads.

You're also ignoring that Apple was working with a much lower clock rate and much lower core power than AMD. Its much easier to add 10% frequency when you're starting at 3.5 GHz than it is when you're starting at 6.

 

[Io Magnesso]

There is no dream story in this world
There's no such thing as convenient

Even Apple accepts a certain amount of power consumption to improve performance…

 

[Io Magnesso]

You're assuming they got that all from process, and none from design or from deciding to use more power in single core loads.

You're also ignoring that Apple was working with a much lower clock rate and much lower core power than AMD. Its much easier to add 10% frequency when you're starting at 3.5 GHz than it is when you're starting at 6.

Do you see 6GHz as the normal operating frequency?

 

[Doug S]

Do you see 6GHz as the normal operating frequency?

There's some binning in that number that Apple isn't doing, so maybe 5.5 GHz would be a fairer comparison to the 3.5 Apple was operating M2 at before they went to N3B. Either way, there is a HUGE gap in the frequency range in which AMD and Apple chips are operating. It is shrinking, but I stand by my statement that attributing the gains on Apple's side to "process" and assuming AMD should have the same percentage gains from "process" is folly.

 

[MS_AT]

I will add that the ~40% for Zen5 was true for pure heavy AVX-512 only, but it had some truth to it.

Well, my FFT routines got 90% when running hot from cache

EPYC Zen 5 comes nowhere near +40% vs EPYC Zen 4, in SPECint or SPECfp overall, so it doesnt matter.

Single core? X cores vs X cores or socket vs socket? Or any of those?

from bigger L3 slightly lower memory latency

How does victim cache lower memory latency?

why wouldn't you expect a lower latency?

In which part of the hierarchy you would expect to see lower latency and what would be the reason for it?

N4P 9950X @ 5.7GHz uses ~35 watts

Core alone or core + IOD?

 

[Joe NYC]

How does victim cache lower memory latency?

That meant to apply to the other feature of Zen 6 which is the new interconnect between IOD and CCD and potentially faster memory support.

But for single thread apps, 1.5x size of L3 will reduce cache misses, which is its way of reducing latency.

 

[igor_kavinski]

7GHz would improve the perception of Zen 6% a bit

Yes indeed. Perception would change to Zen 7%

 

[Det0x]

The price you pay for minimal IPC gains. This is why @gdansk is right while increasing clock is the easiest way to increase performance of a core it comes at the cost of power consumption.

View attachment 127064
N4P 9950X @ 5.7GHz uses ~35 watts. Will AMD maintain this level of power consumption at >6.0GHz?

A single PBO boosting Zen5 core @ ~5.85ghz uses 11.1w average in Cinebench R23 ST according to hwinfo

Totalt PPT is ~50w

 

[511]

A single PBO boosting Zen5 core @ ~5.85ghz uses 12.2w average in Cinebench R23 ST according to hwinfo
View attachment 127072

Can you try using Geekbench ST?

 

[Josh128]

Intel isn’t the basis of good cpu design anymore. Their P core are horrible. Anyway, yes AMD has good engineers. 6GHz isn’t some mythical barrier that can’t be broken.

This is a super cope. Intel achieved 6.2 GHz off the shelf boost with their ultra tweaked for frequency 10nm process, ultra binned CPUs, then failed to do so on a far superior 3nm node. AMD failed to move the frequency bar from Zen 4 to Zen 5 despite a tweaked node. There are physical and electrical barriers to frequency increases at a certain point-- inductive, capacitive and transistor switching speed barriers. All these have to line up perfectly to reliably increase speeds. Things become exponentially more difficult as speed increases, its not a linear relation.

 

[Det0x]

Can you try using Geekbench ST?

GB6 locked with affinity to a single Z5 X3D core with SMT enabled
HWinfo cant keep up with such a bursty workload, so average power usage it out the window, but max draw was ~15.7w with 2 threads enabled on a single core accoring to hwinfo

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 9950X3D processor.

browser.geekbench.com

 

[Josh128]

? Apple got a 10% frequency increase from N3B

Are you really comparing achieving a 4GHz clock to a 6.5GHz clock? Apple silicon is power, size, and heat limited, not frequency limited. Node increases help all those things.

 

[Josh128]

The PS3 emulator got a 30% improvement from avx-512.

Source? Ive seen no such evidence.

 

[511]

GB6 locked with affinity to a single Z5 X3D core with SMT enabled
HWinfo cant keep up with such a bursty workload, so average power usage it out the window, but max draw was ~15.7w with 2 threads enabled on a single core accoring to hwinfo

ASUS System Product Name - Geekbench

Benchmark results for an ASUS System Product Name with an AMD Ryzen 9 9950X3D processor.

browser.geekbench.com

View attachment 127078
View attachment 127077

Thanks so looks like 12-13W/Core not bad tbh

 

[Josh128]

WCCFKEK now regurgitating this speculation page as rumor for Zen 6 node, lol.

AMD to Rely on TSMC’s N2P Process for Most Zen 6 Products, Lower-End Laptop SKUs Likely to Use N3P

TSMC's N2P process will supercharge AMD's Zen 6 products, as Team Red appears to combine an architectural leap and node advancements.

wccftech.com

 

[LightningZ71]

Source? Ive seen no such evidence.

Tom's hardware story on RPCS3 author's own statement about adding AVX-512 and the perf boost it brought...

 

[igor_kavinski]

Yes I got these numbers from my bottom of my ass just like MLID did.

Be better than MLID, my dude!

You guys will never believe this BUT

I have a guy in my office who recently joined.

He is the SPLITTING image of MLID.

Someone told him that I'm some sort of guru.

So he was visiting my seat every day at like 1 or 2 PM but I arrive for work at 3 PM so thankfully I missed him.

He was very enthusiastic about learning stuff from me. But his job is totally different from mine. He's a project officer. Still, his persistence was pretty annoying. He finally gave up and now probably even soured by my reluctance to take him under my wing and teach him the ABC of stuff I do.

I imagine MLID is the same. Relentless bastard who stands outside the home of whoever sources he knows, looking like a pathetic weasel in the rain and they throw random tidbits at him just so they can see his stupid wagging tail disappear far off into the distance, fading as he hurries along.

 

[Win2012R2]

Unfortunately, MLID is a dangerous person who spreads hoaxes

He does it for money. There is a word for this and it starts with f...

 

[511]

He does it for money. There is a word for this and it starts with f...

the entire tech media do it for money gone were the time when it was passion for people

 

[Win2012R2]

the entire tech media do it for money gone were the time when it was passion for people

Doing honest work for money is ok, but when one spreads hoaxes (or lies) for money that become fr....

Some people datamine to find clues as to what will happen - this is fair play, but he claims to be using sources inside orgs to leak confidential info that can cause serious damage to companies, there is no public interest in knowing 18 months too early what frequency will be in a product that takes years to design. This isn't journalism.

It's not like he does it just a couple of months before release either, frankly I am amazed AMD did not take him to court - Sony did the right thing to slap him.

I guess he looks superior to that RGT guy who just spews total BS with a super annoying voice to boot.

 

[511]

i can't remember a single modern tech tuber doing it for passion maybe

 

[igor_kavinski]

Sony did the right thing to slap him.

Link please. I would love to read that!

 

[Markfw]

Tom's hardware story on RPCS3 author's own statement about adding AVX-512 and the perf boost it brought...

Only stated 30% for PS3 emulator, no other numbers