Discussion Apple Silicon SoC thread

[senttoschool]

Mobile Zen never consumes >20W for pure ST loads. Desktop does, mobile is usually around 15W. It also comes with lower peak performance though, so that's to be expected.

For iso-performance Zen 4 will likely be significantly more competitive with Apple's current end cores in performance/W. A lot of power is used for that last roughly 15% performance. To put things into perspective, Zen 4 uses similar levels of power for 4.5GHz as Zen 3 uses for 3.5GHz. That being said, AMD still have a significant disadvantage on power coming from their uncore by comparison however - what I said only applies to looking at the core power only.

I was referring to Package power because package includes powering on things like L3 cache which is used by the CPU core.



For 1T, it seems like the package power is 24w. Core itself is 12.3w.

Here's my M1 Pro CPU wattage data during ST portion of Geekbench5.

Wattage info is shown every 1 second.

We can see that the M1 Pro stays below 2w for most the ST test. It peaked at 5w for 1 second.



I'm not an expert in CPU testing nor do I have a Zen3 or M2 laptop. But at the very least, it seems like M1 is roughly 3-6x more efficient than Zen3 in ST.

 

[senttoschool]

Here's the data for the MT portion of Geekbench5.

It peaks at 35w. Just a ballpark estimate but M1 Pro seems like it's roughly 2-4x more efficient than 6900HS in MT. The 6900HS closes the gap somewhat in MT efficiency because AMD chips tend to excel in MT.

 

[Eug]

Highest result for the M2 Max
View attachment 72532

Mac14,6 - Geekbench Browser

Benchmark results for a Mac14,6 with an Apple M2 Max processor.

browser.geekbench.com

So are people thinking it’s still N5P?

 

[MadRat]

Apple can engineer around their own OS. Neither Intel or AMD enjoy that luxury because they are targeted at a more generalized OS market. If Apple wasn't winning the efficiency battle it would have been news. Otherwise its just a bad argument. Apple cherry picked their market, concentrating on high end products that are not for the general masses. If Intel and AMD were truly getting harmed by Apple releases, they would have sued to unlock Apple's OS to their processors. The general masses can afford Intel and AMD, and Apple would no longer have their niche. But they don't compete for the same markets so its a moot argument.

 

[richardskrad]

-

 

[senttoschool]

Apple cherry picked their market, concentrating on high end products that are not for the general masses.

Do you consider the Macbook Air, at $799, as not mass market? It's literally the best selling laptop in the world.



FYI, both Intel and AMD design highest-end CPUs first such as the 7950x or 13900K, then cut them down for low end.

 

[poke01]

Apple can engineer around their own OS. Neither Intel or AMD enjoy that luxury because they are targeted at a more generalized OS market. If Apple wasn't winning the efficiency battle it would have been news. Otherwise its just a bad argument. Apple cherry picked their market, concentrating on high end products that are not for the general masses. If Intel and AMD were truly getting harmed by Apple releases, they would have sued to unlock Apple's OS to their processors. The general masses can afford Intel and AMD, and Apple would no longer have their niche. But they don't compete for the same markets so its a moot argument.

Can't Intel and AMD use Linux and focus on that. After all Linux is open source. But Linux is very fragmented.

 

[richardskrad]

There are Windows laptops in the same price segment as all the different MacBooks. If Apple has the superior chips, it should be acknowledged because it shames and lights a fire under Intel And AMD.

One can make the argument that Intel and AMD don’t care because they are primarily designing for the server market and the laptop/mobile market is insignificant. If that’s the case, then that means expensive (+$1K) Windows laptops are outdated and overpriced tech compared to MacBooks and should not be recommended to consumers.

 

[senttoschool]

One can make the argument that Intel and AMD don’t care because they are primarily designing for the server market and the laptop/mobile market is insignificant. If that’s the case, then that means expensive (+$1K) Windows laptops are outdated and overpriced tech compared to MacBooks and should not be recommended to consumers.

Intel's consumer CPU business is actually bigger than their server business.

 

[moinmoin]

Can't Intel and AMD use Linux and focus on that. After all Linux is open source. But Linux is very fragmented.

Actually from a hardware point of view Linux as in the kernel is very centralized, all hardware support is within the single kernel, and external driver modules from 3rd parties are uncommon (outside the elephant in the room that is Nvidia).

Driver wise Windows is very fragmented, Linux not really.

 

[DrMrLordX]

Apple can engineer around their own OS. Neither Intel or AMD enjoy that luxury because they are targeted at a more generalized OS market. If Apple wasn't winning the efficiency battle it would have been news. Otherwise its just a bad argument. Apple cherry picked their market, concentrating on high end products that are not for the general masses. If Intel and AMD were truly getting harmed by Apple releases, they would have sued to unlock Apple's OS to their processors. The general masses can afford Intel and AMD, and Apple would no longer have their niche. But they don't compete for the same markets so its a moot argument.

I doubt a lawsuit would prove fruitful; however, Apple can control software better than Intel and AMD, so that does give them some advantages.

 

[FlameTail]

Apple can engineer around their own OS. Neither Intel or AMD enjoy that luxury because they are targeted at a more generalized OS market. If Apple wasn't winning the efficiency battle it would have been news. Otherwise its just a bad argument. Apple cherry picked their market, concentrating on high end products that are not for the general masses. If Intel and AMD were truly getting harmed by Apple releases, they would have sued to unlock Apple's OS to their processors. The general masses can afford Intel and AMD, and Apple would no longer have their niche. But they don't compete for the same markets so its a moot argument.

The fact that Apple has better optimisation does not affect benchmarks do they? Unless Apple lobbied the benchmark makers to 'optimize' for their chips.

Most of the performance/efficiency gains coming from optimisation would be largely limited to the OS and Apple's own apps.

 

[DrMrLordX]

The fact that Apple has better optimisation does not affect benchmarks do they?

Sure, in any application that uses their fixed-function hardware, it makes an enormous difference. Of course those applications may not have direct ports outside of MacOS so make of that what you will.

 

[scineram]

I doubt a lawsuit would prove fruitful; however, Apple can control software better than Intel and AMD, so that does give them some advantages.

Not much to do with software. They just engineered their processors for all day battery life at maximum performance.

 

[gdansk]

Their processors are better at low power but MacOS, even in its Intel days, was good (in my experience) at getting its advertised battery life. They really made it a focus.

 

[Doug S]

Sure, in any application that uses their fixed-function hardware, it makes an enormous difference. Of course those applications may not have direct ports outside of MacOS so make of that what you will.

How is that any different than optimizing for AVX512, or optimizing for Nvidia/AMD GPUs and ignoring optimizations for Apple's delayed rendering pipeline? Why is optimizing for the special sauce Apple gives you somehow wrong in some people's minds, but those same people won't bat an eye over optimizations for stuff only found in the x86 PC world?

Then you can also worry about benchmarks that are particularly sensitive to Apple's large cache sizes and/or outrageous memory bandwidth, but then you might choose one that does so little it is basically measuring clock rate.

Comparing performance based on benchmarks optimized for things only one side has is useful if you want to e.g. know "which will run this particular rendering application I use faster?" It is when people draw unjustified conclusions and claim "system A is faster for rendering" or "system A is faster" because of that one benchmark that may use fixed function hardware, or GPU, or AVX512 or whatever that only one system being compared has that is unjustified.

 

[DrMrLordX]

How is that any different than optimizing for AVX512, or optimizing for Nvidia/AMD GPUs and ignoring optimizations for Apple's delayed rendering pipeline? Why is optimizing for the special sauce Apple gives you somehow wrong in some people's minds, but those same people won't bat an eye over optimizations for stuff only found in the x86 PC world?

You may as well ask why Bitcoin is mined on ASICs rather than AVX512-capable CPUs. Answer that and you are on your path to the truth.

 

[Doug S]

You may as well ask why Bitcoin is mined on ASICs rather than AVX512-capable CPUs. Answer that and you are on your path to the truth.

That's a terrible example. Bitcoin mining isn't an application you run alongside other applications, it is the sole purpose for buying whatever hardware it is running on. Except in rare cases, no one is buying hardware exclusively to do rendering, or run Office, or browse Javascript heavy sites, or run a particular game, or any of the individual benchmarks that people run to try to characterize the performance of a system. That's why people run a bunch of benchmarks, so you can look and say "well I never do rendering or run Office, but I surf the web and play games so I'll ignore the first two categories and pay attention to the last two".

Hopefully all the people buying as well as selling all those bitcoin ASICs are losing their shirts, so that waste of energy can return to the true value of tulip bulbs.

 

[DrMrLordX]

That's a terrible example.

You could classify it as reducto ad absurdum, but it does serve to illustrate the point.

Except in rare cases, no one is buying hardware exclusively to do rendering, or run Office, or browse Javascript heavy sites, or run a particular game, or any of the individual benchmarks that people run to try to characterize the performance of a system.

But Apple buyers do buy hardware exclusively to run Apple software and/or software that has been compiled using Apple's toolsets + Apple's libraries, and conforming to Apple's APIs.

 

[scineram]

So what? No amount of magic compiler flags will make up for orders of magnitude of performace/power difference.

 

[scineram]

So what? No amount of magic compiler flags will make up for orders of magnitude of performace/power difference.

 

[DrMrLordX]

So what? No amount of magic compiler flags will make up for orders of magnitude of performace/power difference.

Company X with no control over compilers, APIs, libraries, and OS introduces fixed-function hardware as an add-in card to accelerate Y common workload with fairly common repeating algorithms/functions that can easily be modeled in an ASIC. Company X gains no traction and, if lucky, is bought out by a larger competitor.

Apple with control over compilers, APIs, libraries, and OS introduces fixed-function hardware as part of the SoC package to accelerate Y common workload with fairly common repeating algorithms/functions that can easily be modeled in an ASIC. Everyone wanting to sell software on Apple's platform complies with Apple's dictates and easily makes use of this hardware due to the supplied toolset and backend. The hardware is successful, and many end users enjoy benefits from it.

 

[scineram]

Endless meaningless excuses for taking 3 times the power or more for similar Geekbench ST scores.

 

[DrMrLordX]

Endless meaningless excuses for taking 3 times the power or more for similar Geekbench ST scores.

Yeah because that's the end-all, be-all of computing benchmarks.

Meanwhile:

Apple M1 Performance On Linux: Benchmarks Better Than Expected For Its Alpha State Review - Phoronix

www.phoronix.com

MacMini M1 getting spanked by a 5700G in an environment where software wasn't optimized for all the fun bits in the M1.

 

[scineram]

As shown above over 2000 GB points, in the ballpark of desktop Zen 4 and Raptor, for a fraction of power with M2. Definitely the most-all for the vast majority of PC market that are laptops. But even for desktop users is quite based.
The phoronix bench is also kawaii with M1 having only 4 perf cores, when we already have M1 Pro and up and soon M2 Pro.