Apple Announces Its New M2 Processor

[WannaBeOCer]

Yeah this is the bottom line. The M series is nice in theory, and only great in practice if you can handle being stuck on OSX.

Sometimes thats fine, but sometimes its not.

Maybe it will work on linux (and by extension Windows through WINE) some day, but it does not work now, it will not work tomorrow, and it probably wont be full speed for a long time.

What's wrong with being stuck on macOS aside from gaming? Which I see them making a huge impact this year since their architecture is similar between all their products. They even announced MetalFX Upscaling. They're way ahead of everyone else when it comes to ARM based apps and seems like they're doing fine with hardware. Unlike Windows/Linux all their apps are 64-bit, their HDR implementation is practically perfect.

[tunnelblick]

The M2 MBA is, again, fan-less. What else is there to say? Wake me up when AMD or Intel come up with a chip that can be cooled with a heat pipe and delivers the same performance the M1/2 offers.
You can dislike Apple and their OS or their philosophy or whatever but their CPUs are great.

[sinepgib]

What's wrong with being stuck on macOS aside from gaming?

Maybe it's just not the OS you'd prefer to use, for whatever reason. A few reasons one may not want MacOS that come to mind:
- Preference for open source;
- Having to work with Docker and preferring a native experience;
- Being used to Windows and Windows software;
- Having a library of games unlikely to be ported.

Now, nobody's saying you personally shouldn't use it or like it, but for a lot of people MacOS is not an appealing option, even while the hardware may be.

[HEL88]

X86 chips still pay the price despite all the instruction caching they claim. There's no free lunch for having a bad ISA. That caching is of limited size, consumes massive amounts of die area in addition to the decoding circuitry, and the ISA still imposes a low limit on how quickly you can decode *new* code while following program execution. .

So, ARM is bad ISA too, because Cortex A77, A78, X1, X2, Neoverse-V1, Neoverse-V2, Neoverse-N2, every top power ARM procesor since 2019 have uOP cache like x86.

A78, X1, X2, V1, V2 have big 3000 uOP like Zen2/3 other (A77, N2) have smaller 1500 uOP.

[qarium]

Fair enough.



Pure Python is single threaded. I didn't check the source code to see if any of the C modules in the standard library does release the GIL, but my guess is also no. Stuff like NumPy and what not is of course parallel because those make a point of releasing the GIL.
Kernel calls are not multicore from the POV of Python: regardless of if it blocks or is non-blocking, it's a different thread that will take its turn to run, but when it does it will take the GIL. Essentially, pure Python can only parallelize IO.



Indeed. But the system is not an application, is the whole system. I'm discussing whether most single applications do use multicore properly, which was what you said.



Sorry, but what do you suppose my theory is, exactly? My only "theory" is that to claim most apps make proper use of all cores is something that needs a lot of data to backup. I never contradicted, and in fact I asserted, that multicore performance is by today's standards absolutely more important than single core performance. Single core performance would only matter if your "system" was actually running a single process at a time, a-la DOS.
In fact, if any criticism to the world is to be emitted from what I said, is mostly that I don't think everyone does a good job of parallelization, certainly not that we should go back to optimize for the single core and single process past.

I'm not quoting the rest of the post because you're clearly not reading what I'm saying anyway. The numbers I ask for are not performance numbers, but statistics about multicore applications and how they scale. It was never an argument against whether or not it was better to optimize hardware for single or multiple concurrent tasks, as I already stated it's obvious the second case is what's desirable in multi-tasking systems (i.e. everything but the smallest embedded).

you say modern systems does not "does a good job of parallelization" and this is right if you consider the fact that there are single socket 128core systems and if you use todays software stack on these 128core cpus then the result on every day tasks is that it can not utilize the 128cores. but thats not the point at all because it is pointless the big part of the market do not buy 128core cpus.

you want these numbers: "but statistics about multicore applications and how they scale." you already have it without you know it.

the same reason why the main part of the market is not buy 128core cpus is the same reason why they buy 8core cpus.

they buy 8core cpus because the today software stack is scaling on these 8 core cpus. they do not even buy 12core or 16core cpus because this does not """scale"""

"Kernel calls are not multicore from the POV of Python:"

whatever python calls from the kernel the kernel does multicore internally.

"Essentially, pure Python can only parallelize IO."

yeah right first you say python can not do multicore and now you claim python can do multicore for IQ

your system run multible python programms in parallel... just in case you miss this point.

"Sorry, but what do you suppose my theory is, exactly?"

in my point of view your theory is single core performance is all you need because of this you buy "overpowered single core cpu" well because this is no longer in the market you buy any cpu with the highest single core performance in the market.

the market alone proof that single core performance is not longer relevant because the """average""" consumer buy 8cores because he can untilize 8cores... he does not buy 6core because of the lag of cores and he does not buy 12/16core because to many threats needed to utilize the cpu...

a 6 core is cheaper than a 8 core so why do people buy the 8core if only single core speed matters ?

"in fact I asserted, that multicore performance is by today's standards absolutely more important than single core performance."

you say this yes but you claim otherwise you claim your python app is only singlecore...

so how can multicore performance be more important if your python app is only single core ?

in reality it is nonesense then as soon as your system runs 2 different python tasks who do not need to sync you are already in the multicore world

[tildearrow]

dude honestly you mix up 2 different things... to have a "monopoly" and to have the the crown in a product segment because you have the best product.
apple maybe has the best product in the low-power high-performance CPU market field what gives them a crown at this point.

but at no point makes this apple have a monopole... monopole in my point of view is something what FORCE you to buy it.

without renounce many stuff you can not ignore windows or intel/amd X86_64 best example is valve steam you have 3000 games on steam OS but you have like 8000 games on windows means you can not ignore windows as a gamer if you do not want to renouce large part of the market.

the same at X86_64 without such an CPU and you maybe have an ARM64bit cpu your basket of games goes from 20 000 to maybe 1000....

so apple has the best low-power high-performance CPU on the market but this does not give them any monopole

i have an ARM smartphone and 86_64 desktop and i have ZERO apple products..-. because they have zero monopole

i even hate apple because of my personal history as a child was forced to use an apple macintosh and all the other kids used pirated software on pc but i have to admit apple has """zero monopole"""

in business and job you are maybe forced to use microsoft office but honestly apple has zero product like microsoft office to force you into their stuff.

and just because a apple product is better does not magically turn it into a monopole better does not mean monopole...

its the opposite companies who have monopole can force you to use a inferior product for example microsoft office is clearly a inferior product but the high marketshare and monopole position makes them powerfull even so powerfull they can force inferior products on the market.

apple has products who are better... but they have zero product with a monopole means they lag the high marketshare to give them the opportunity to force inferior product on people.

"and had to reverse engineer the architecture to no end (e.g. Asahi Linux) just so that they could run a better operating system."

apple locks out everything on their iphones but apple allows 'Asahi linux on apple M1...

in the past evil monopolist microsoft did try to lock out linux from windows laptops just read halloween documents they did multibble stradegies to make this happen.

"Ampere will never look at us (instead exclusively focusing on server) and so we will be stuck to either:"

what a bullshit talk... they sell 128core versions but you can also buy 64core versions or 32core versions or 16core versions
if you pay the price you can already use it on workstation and desktop.

you can not do it because Wintel monopole most games and closed source products do not run on ARM CPU and also microsoft does not really work good on these ARM systems to.

but it is clear it is not Ampere who does sapotage you it is intel and microsoft who do sapotage you.

no one stops you to even put a 16 or 32core ampere cpu into a laptop but still outside of server who linux dominsate the evil intel and microsoft monopole will make you suffer.

we can say apple is the only company because of their lock-in stradgey is the only company who is able to break out of this intel and microsoft monopole

and we should be happy about this. because we suffer from microsoft/intel monopole for like 40 years now.

so stop talking bullshit about words you do not unetrstand like "monopole" an product who is better is plain and simple still not a monopole.

I've rephrased my comment (Phoronix doesn't seem to like swearing on edit messages though).

[AdrianBc]

Acceptable baseline performance?

Apple M1 literally destroys top Intel and AMD CPUs in terms of absolute performance (per core <10W power envelope vs. e.g. ~25-30W for ADL/Zen3) and power efficiency (MT performance).

And their unified memory architecture? The PC can only dream of it. XBox/PS have it though but they are not PCs.

Apple M1 and M2 destroy the competition in terms of performance per watt or per joule, not in terms of absolute performance.

Even a desktop Zen 3 is a little faster than an Apple M1 in most applications and much faster in a few applications (e.g. computations with large numbers).


There is no need of other proof than the actual Apple presentation, which has acknowledged that the new M2 barely reaches 87% of the performance of a relatively slow Alder Lake version, the 28 W 12-core i7-1260P. However, it does this while consuming only 1/4 of the power consumed by the Intel CPU, so yes, it destroys the competition in terms of energy efficiency, even if the absolute performance is lower.

Slower than Alder Lake P means much slower than Alder Lake H, Alder Lake HX or Alder Lake S. Moreover, for most of its life, M2 will not compete with Alder Lake, but with Raptor Lake. If it is slower than Alder Lake, it will be much slower than Raptor Lake and likely also slower than Zen 4.

Of course, there is no doubt that M2 will remain far better in energy efficiency than any of those, even if it will not reach their absolute performance.

[Keats]

What about idle? Majority of desktop/server/workstation x86 motherboards idle at 5-25W, whereas M1 and Raspberry Pi idle at <5W.

Majority of desktops/servers/workstations aren't very concerned with idle power consumption (indeed, for servers that's probably near the very bottom of a long list of priorities). If you want low idle power consumption, you should probably look at SFF computers that use laptop parts.

[AdrianBc]

This post argues that's not the case and that the reason M1 manages to do so well is a fundamental limitation in x86 ISA chips' ability to parallelize instruction decode for a CISC architecture with variable width opcodes. (In addition to other things like "SoCs have certain advantages and there aren't any x86 SoCs in that market segment yet".)

That post makes some suppositions, but it does not prove anything. For that it would have to measure a real computer.

An Apple CPU does about 3/2 more work per clock cycle, while using clock frequencies of only about 2/3 of those used by the competition.

For CPUs made with the same CMOS process, this approach is guaranteed to use much less power. Moreover, the Apple CPUs are also made with a better process, which reduces even more the power consumption.


There is no doubt that if the power consumption of an Intel or AMD CPU will be reduced to the same value as for an Apple CPU, the Intel or AMD CPU will be significantly slower.

However, I have not seen any published results of such a measurement, so we do not know exactly how much slower.

What is certain is that the difference in performance at equal power will be much less impressive than the difference in power at equal performance.

For example, the new Apple M2 has 87% of the performance of a 12-core Alder Lake at 25% of the power.

Lowering the power of the Alder Lake until the performance matches the Apple M2 would reduce the power ratio from 4 to somewhat less, maybe between 3.5 and 3.8.

Lowering further the power until it matches the Apple M2 power will lower the Alder Lake performance, but in a ratio much smaller than the 3.5 to 3.8 ratio in power.

It is likely that at equal power Alder Lake will still have a higher clock frequency than the Apple M2 and it will be only about 1.3 to 1.5 times slower.


An Alder Lake or a Zen 3 with the power consumption reduced to that of an Apple CPU, would be slower than the Apple CPU, but still faster than a top-speed Skylake CPU of a few years ago, so good enough for most tasks.

[carewolf]

Apple M1 literally destroys top Intel and AMD CPUs in terms of absolute performance (per core <10W power envelope vs. e.g. ~25-30W for ADL/Zen3) and power efficiency (MT performance).

No, you are linking to an M1Pro review, and you added so many constraints on that statement, it is cherry-picked.