Apple's A17 Pro Within 10% of i9-13900K, 7950X in Single-Core Performance
tomshardware.com
tomshardware.com
If they can drive the same increase with an M3 chip based on the A17 Pro microarchitecture, a single-core score of 3500 is within reach. For a laptop.
Comparison: the i9 13900KS, the current stock single-core performance champion, can push 3100 with excellent cooling; and it reaches this blistering performance (~10% faster than an M2 Ultra) by, singularly, drawing upward of 220 watts from the wall [1]. The Ryzen 7900x isn't far behind, in both metrics. An entire M2 Max SoC peaks at around 90 watts, and there's evidence that the CPU in isolation draws closer to 35 watts [2].
It cannot be understated how far ahead Apple is.
[1] https://www.tomshardware.com/reviews/intel-core-i9-13900ks-c...
[2] https://www.notebookcheck.net/Apple-M2-Max-Processor-Benchma...
If you run Geekbench 5 or 6 on an M1, the total package power ranges 0.3w - 5w, mostly staying below 3w. Intel and AMD can use 10x or more power to achieve similar or lower Geekbench scores.
https://www.notebookcheck.net/AMD-Ryzen-9-7940HS-analysis-Ze...
Current laptop AMD chips have about the same efficiency as M2 Pro, when throttled to similar wattage (see the section about 35W). Instead they default to more power and being more powerful, while being vastly cheaper (M2 Pro Mac mini with 32gb of ram and 2tb of SSD costs what?).
So yeah, for a laptop chip apple is ahead, but it is pretty easy to overstate by how much.
M2 Pro @ 141 points/watt Ryzen 9 7940HS @ 47 points/watt
It's kind of depressing that the most advanced mobile CPU technology (that can only be fabbed by a single company in a far-away country) is monopolized by a company that heavily restricts the software you can run on it, and will never sell it to me with documentation and let me plug it onto my own motherboard--which is ironically how Apple first started, by buying and making full use of the MOS 6502. But the days of the garage hacker getting their hands on cutting-edge parts are long gone.
It’s bad that they can gatekeep like this, I agree. I should just be able to put whatever on the hardware I buy.
But the expense in practical terms, immediate pragmatic terms, is negligible compared to what it unlocks.
The promise of general purpose computing and open source as well is that it empowers all kinds of users.
Oh how badly I want to tweak little things about my iPhone, but can’t because the software is locked down.
If you want to do toy programming on a budget there are a million arduino-like things out there.
I recently saw calculations on the price of say, just iOS alone. It being equivalent to multiple Manhattan projects. Regardless of how accurate that estimate is, the general point remains.
Do I think Apple should make developing for non-distribution on the iPhone free? Yes. Do I think there is an argument that is almost a moral obligation? Possibly.
But what you get for like 30 cents a day with a developer account is mind boggling. I mean, that was the initial point wasn’t it? These are insanely capable and meticulously engineered pieces of technology.
Edit: I can even argue it is an ethical obligation that Apple should allow all users to install non App Store apps if they explicitly so choose. And I have argued that in the past.
But I separate this from the evaluation of the value proposition offered by an Apple Developer account.
Now it’s a whole other discussion of if that’s a good thing or not. But I personally like to believe that everyone should be able to develop as a hobbyist. The commercialization of software is often at odd with it’s users. Not so much when people are doing things for themselves and for fun.
Just my 2¢
The limitation is that you will have to revalidate weekly (which can be automated).
Does that mean that x86 has a specific purpose now and is no longer a general purpose CPU?
Oh, yes, you can.
The following examples of
– Transport triggered architecture
– Dataflow architecture
– Optical and quantum computing
represent viable, general purpose (not general availability), albeit experimental, CPU's without instruction sets. That is, none of them have anything similar to «movq $1, %r0» or «add.w %r2, %r0, %r1».
FPGA's are another and a more conventional example of a general purpose (even if specialised) CPU without an instruction set.
Sure, apple suck at sideloading, hopefully the EU ruling will make it a possibility, but then say that instead.
That sounds like these "technically, a taco is a sandwich" discussions. Maybe, but was that actually the question?
In the context of CPUs i'm with you (still tinkering with things like 6502s).
On the other hand, i just deployed Llama2 70B (with the help of [0]) on my home PC and it was fascinating, when it wrote it's first response.
Sure the chips are fast when they run the stuff Apple designed for it, but they don’t seem any faster to me in my work. I will give them efficiency though. I love not having to carry the charge brick everywhere I go.
Whenever I open IntelliJ on it I can feel the temperature of the room increase. The Intel Macs were terrible for many years.
Docker sucks because it is a linux-ism, not any other reason. And I think most people would disagree with your notion, it really does run circles around basically every other laptop.
Both have apple silicon builds now, but it and a lot of other software built for Intel ran at similar performance or better (compared to Intel MacBook / Thinkpad).
I understand why people hate apple, but I've never heard anyone say this (what op said) who has had similar setups. It works a hell of a lot better than my X1 Carbon did.
It'd be worth checking that you're not using an Intel version of WebStorm. I've personally found IntelliJ (largely the same thing as WebStorm) on the M1 to be _extremely_ fast, certainly leaps and bounds ahead of the old Intel Macs. That's the native version, though; via Rosetta it is absolutely painful (the JVM is close to a worst-case for Rosetta).
Apple uses the same CPU/GPU core designs for handheld, tablet, laptop, and desktop.
Also, even without that there is ish which can run most x86 software.
Is the idea that mobile workloads are spiky so it can boost to full speed to render a webpage for example then go back to sleep? Are there / can there be any applications that sustain high clock speeds?
This is why modern consumer processors have boost modes followed by thermal throttling.
And if the implication is that they'll port those iOS games to Mac, then I wouldn't count on it if it takes even a single iota of extra effort on their end. There are more Linux gamers than Mac gamers, and yet game devs still can't be arsed to test their games on Steam Deck.
Tale as old as time.
Apple has historically been terrible about committing to gaming which has led to game developers ignoring the platform. I do hope that Apple actually stays the course this time around and with the M* series of chips they've raised the floor for what game developers can count on having available.
I really thought they would have made a bigger play for gaming on the Apple TV (The hardware, not the app, not the service, I still can't get over that...), they pushed a little at the start but with most Apple gaming initiatives (aside from mobile) it seems to have fizzled out.
The games will come to all Apple Silicon devices as the architectures are very similar if not the same (M1 iPad vs M1 Mac). Apple also released the Game Porting Toolkit and I can already play games at decent framerates with Whisky [0] which utilizes GPTK. Keep in mind this is basically beta software that's not even for the end user to use but as a developer tool to help port their games over.
And obviously the games when on iOS won't be played with touch controls, it's for playing with a Bluetooth controller.
I think this would be the base of the future M4.
This is also why on iOS, apps that become known for abusing or mismanaging background processing are penalized and have their background processes run less often. Even just a couple of processes waffling around instead of completing their tasks and quitting can keep the system from idling and have a big negative impact.
This is why numbers like this don’t matter when you say the performance is within 10%
Why does GHz matter? Performance should be measured as a black box, with the metric being the time it takes to accomplishing some task.
Maybe it can run faster given better cooling and less aggressive power-saving. But I'd be surprised if it can be made to clock much higher, else Apple would have done it already. Mobile CPUs already have peak performance well above what they can sustain. An even faster turbo would be used more rarely but would still be useful.
of course there’s more than just thermals. The CPU has to actually be designed to run at a high clock.
I could be wrong but my understanding is that all their CPUs share a good amount of commonality and the major differences are more around packaging (IO connectors etc)
It correlates less directly between families, smaller process nodes tend to allow for high frequencies at the same power consumption.
Is the gap expanding at this point?
[1] https://beebom.com/snapdragon-8-gen-2-vs-apple-a16-bionic-be....
> Apple’s A16 Bionic chipset scored 1879.
Both on the same 4nm TSMC process.
Antutu CPU - A16 248335, SnapDragon 8 Gen2 268542
Antutu GPU - A16 394336, SnapDragon 8 Gen 2 580101
If You look at Geekbench 5 scores for A16 and SD8G2, A16 is ahead by less than 10%.
power efficiency aside (which admittedly is non-trivial), this is not that big a deal or particularly interesting, imo.
Apple’s lead is almost unfathomable.
I'm not sure how realistic this is but I think I'd love it.