Let's see if I can turn this into an ASCII table and have it survive HN's reformatting.

    +------+------------------+--------------+----------+----------------+-------------------+-------------------+---------------------------+
    | Chip | Process          | CPU Cores    | GPU      | Neural Engine  | Memory Bandwidth  | Unified Memory    | Geekbench6 (Single/Multi) |
    +------+------------------+--------------+----------+----------------+-------------------+-------------------+---------------------------+
    | M1   | 5 nm             | 8 (4P+4E)    | 7–8      | 16-core Neural | 68.25 GB/s        | 16 GB             | ~2346 / 8346              |
    | M2   | 5 nm (G2)        | 8 (4P+4E)    | 8–10     | 16-core Neural | 100 GB/s          | 24 GB             | ~2586 / 9672              |
    | M3   | 3 nm (first-gen) | 8 (4P+4E)    | 8–10     | 16-core Neural | 100 GB/s          | 24 GB             | ~2965 / 11565             |
    | M4   | 3 nm (second-gen)| 10 (4P+6E)   | 8–10     | 16-core Neural | 120 GB/s          | 32 GB             | ~3822 / 15031             |
    | M5   | 3 nm (third-gen) | 10 (4P+6E)   | 10       | 16-core Neural | 153 GB/s          | up to 32 GB       | ~4133 / 15437 (9-core)    |
    +------+------------------+--------------+----------+----------------+-------------------+-------------------+---------------------------+

The step down from 32GB to 24GB of unified memory is interesting. Theories? Perhaps they decided M4 allowed too much memory in the standard chip and they want to create a larger differential with Pro/Max chips?

Update: I am thinking the 24GB for M5 is a typo. I see on Apple's site the 14 inch MBP can be configured optionally with 32GB of RAM.

Amazing. My M3Max is going to look like a paper-weight very soon. And that's fine by me. When I get an M6 or M7Max to replace it it'll be amazing.

The multi-core geekbench score for the M5 is the 9 core version iirc. The 10 core score isn't out yet as far as I know.

Did TSMC 2nm slip to next year, or was it always planned to be 2026?

Thank you! Since this is the top rated comment, can you also add M1 and M2 as well?

Does this mean the M5 is serious as fast as my intel 13900 cpu?

That's a lot of memory bandwidth. Kinda surprised geekbench doesn't benefit more from the fatter pipe.

Serious questions. How is Asahi these days? Is it ready as a daily driver? Is it getting support from Apple or are they hostile to it? Are there missing features? And can I run KDE on it?

And the fastest M4 max was already fastest single and multicore CPU by a decent margin, while the fastest non-Apple CPU was only specialized for single or multi.

Interesting to see that over 5 years (M1 was 2020), the benchmark performance has not quite doubled. Is this an indictment of Moore's law, or just Apple over-speccing the M1 and slowly decreasing that over time?

Thank you. Looking at replacing an Intel MacBook Air, I hope there are price drops on the "outdated" M4s (although an M2 phased out early this year would do well enough...)

They're going to have a hard time selling the M5 when compared to the M4 Pro. Geekbench for that chip is 3843/22332, which is slightly slower for single core but better for multi, but also has thunderbolt 5 instead of 4.

> Tahoe however makes my M1 Air feel sluggish doing the exact same tasks ive been last couple of years.

I have a work provided M2 Pro with 32GB of RAM. After the Tahoe upgrade it feels like one of the sluggish PCs at the house. It is the only one that I can see the mouse teleporting sometimes when I move it fast. This is after disabling transparency in Accessibility settings mind you, it was even worse before.

There are so many software related things that drive me absolutely loony with Apple right now.

* My iPhone as a remote for my Apple TV has randomly stopped deciding it can control the volume - despite the "Now Playing" UI offering an audio control that works.

There auth screens drive me crazy:

* Why cannot I not punch in a password while Face ID is working? If I'm skiing, I know Face ID isn't gong to work, stop making me wait.

* Likewise, on Apple TV the parental control input requires me to explicitly choose to enter a Pin Code. Why? Just show me the Pin Code screen. If I can approve from my device, I will.

  * Similarly, if I use my phone as a remote, why do I need to manually click out of the remote to get to the parental control approval screen. I'm literally using my phone. Just auto-approve.

It's incredible what the hardware teams at Apple have been doing. I imagine they also feel let down by the software that's driving these beasts. It's as if they're 2 completely different companies.

>The M5 MacBook Pro still gets the Broadcom WiFi chip but the M5 iPad Pros get the N1 and C1X (Sweet).

Is that good? Their cellular modems have been terrible. I'll reserve judgement until trying one out.

>The M1 itself is so powerful

I think this is a bit of a fallacy. Apple Silicon is great for the power consumption to power ratio, but something like a Ryzen 9 7945HX can do 3x more work than an M1 Max. And a non-laptop chip, like an Intel Core Ultra 7 265k can do 3.5x.

I really wish apple sold the Mx to others like Lenovo.

I would love to se a ThinkPad with an M5 running Linux.

The Broadcom WiFi support 320Mhz while N1 is stuck with 160Mhz. There were report of N1 not supporting 4096 QAM as well but I didn't check.

Yeah I love my M1 iPad Pro. But the "liquid glass" update made it feel slower. Really only the 'unlock' feels slower, once I'm using it it's fine. But it's slightly annoying and does make me want to update this year to the m5.

But it's a glorified Kindle and YouTube box, so I'm hesitating a little bit.

I’m still daily driving my M1 Max and have no reason to upgrade for a long time. There’s really nothing in my workflow that could be markedly improved performance wise. There’s only thing is maybe more ram as the need for that keeps growing - I’m isn’t just under 30 when running a bunch of containers.

My M1 Air got very sluggish after upgrading to Tahoe but then it started behaving normally after a couple of days. Hopefully, you'll experience the same soon.

> The M5 MacBook Pro still gets the Broadcom WiFi chip but the M5 iPad Pros get the N1 and C1X (Sweet).

I think many IT departments will be thankful for that as Wifi behaviour can be challenging and hopefully will lower ticket counts.

As a UI/UX nerd, it’s a coin flip on intentionality. I’ve been noticing so many rough edges to Apple’s software when it used to astound. iOS Settings search will flash “No Results” as you begin to type which is comically amateurish. The macOS menu bar control panels can’t be keyboard navigated... It’s just silly.

I’ve been debating making a Tumblr-style blog, something like “dumbapple.com,” to catalogue all the dumb crap I notice.

> Tahoe however makes my M1 Air feel sluggish

Counterpoint: my M1 Pro was a turtle for a few weeks and then stopped doing nonsense in the background and is back to its zippy self. (Still buggy. But that would be true on new hardware, too.)

I would be soo excited if apple split out the hardware and software orgs and moved to make hardware more standardized with macos/ios/etc being just one consumer.

Not going to happen, but I can dream.

I think it's probably a play to get you to upgrade for the new GPU computational power. I _do_ think that what we're seeing (and marketed as AI) will be the future, but I don't think it will look like what we're seeing now. Whatever that future holds will require the upgraded capabilities of these new GPU architectures, and this being a reason for the subtle nudge to upgrade from Apple makes sense to me.

It feels very much like how I imagine someone living in the late 1800's might have felt. The advent of electricity, the advent of cars, but can't predict airplanes, even though they're right around the corner and they'll have likely seen them in their lifetime.

Seems like the software teams are there to simply squander the extra processing power that the hardware teams provide, thus ensuring recurring revenue. I see no good reason to upgrade to Tahoe. I’d have to buy a new computer just so I could power transparencies that I don’t want.

Won't that make Linux support even harder :/

They always release a slowdown update to destroy their older hardware. I don't know why you're even questioning it

“nobody really needs to upgrade that for most things”

Maybe, but for lots of scenarios even M5 could still benefit from being an order of magnitude faster.

AI, dev, some content scenarios, etc…

This needs benchmarks.

Sad if true. I feel my M1 max sluggish too lately. After bragging that this was the longest lived work machine I had and thinking I'm good to wait for M6. This is not good for business, but IMO you need more than raw power to justify upgrades even for professional use - form factor, screen quality, battery, etc.

I think they bet a lot of hardware money on AI capabilities, but failed to deliver the software, so there was no real reason to upgrade because of AI features in the chip (which is literally what they boast on the first line of the announcement - yet nobody cares about making more cute faces)

It's not 100% their fault. Everyone got onto the LLM bandwagon like it's "the thing" so even if they didn't believe it they sill needed something. Except an OS is not a chat interface, and LLMs do suck at stricter things.

Which is harder these days, software or hardware?

>The M1 itself is so powerful that nobody really needs to upgrade that for most things most people do on their computers

a rant on my part, but a computer from 10 years ago would be fine for what most people do on their computer, only for software bloat..

Even the M4 Max MacBook, I tried in the stores were running like shit on Tahoe

Don't kidd yourself: Planned obsolescence is real.

Apple has a higher duty to their shareholders than to their customers.

Not hating on Apple, just stating the hard economic truth.

> ...The <thing I own right now> is so powerful that nobody really needs to upgrade...

I keep hearing this since the Intel 486DX times, and

> Nobody will ever need more than 640K of RAM!

If Tahoe made M1 slower then I am sure it was intentional. Apple had done this in the past and been fined for hundreds of millions in different courts all over the world. So I am pretty sure they continue slowing software down intentionally on older hardware. You can google "batterygate" and you can see for yourself

I'm quite upset I have nobody I know in real life who will appreciate this line.

What does AI performance even mean for intel based mac systems.. The last one was like 5 years ago?

God bless them if so

> A lot of Apple hardware is impressive on paper, but I will never buy a Mac that can't run Linux.

They run Linux actually very well, have you ever tried Parallels or VMware Fusion? Especially Parallels ships with good softwaer drivers for 2d/3d/video acceleration, suspend, and integration into the host OS. If that is not your thing, the new native container solution in Tahoe can run container from dockerhub and co.

> I simply don't want to live in Apple's walled garden.

And what walled garden would that be on macOS? You can install what you want, and there is homebrew at your fingertips with all the open and non-open software you can ask for.

It sounds like Linux works fairly well on Strix Halo, which basically gives Apple a run for their money and stays in the nice x86 land. The M1 and M2 chips were envy-inducing chips from the heavens or whatever, but now that the mortals have caught up I don’t really see the point in worrying about Linux on ARM. X86 remains the present, RISC-V is the future.

I'm still looking for a decent ARM laptop that runs linux well. I have my eye on one from lenovo but linux support is still not the best

M1 and M2 Macs run Asahi Linux very well (but no option for M3,M4,M5 yet)

Isn’t to core of MacOS derived from Unix?

I came to chime in. I have hardware that apple chooses to willfully upsell me on repairing and $1500 for $35 keyboard repair. Apple as a company is still terrible at recycling and manufacturing obseletion. It is also a walled garden with no choice as to what you can do on your machines.

honestly, computing speed doesn't matter that much anymore

I mean as long as the law of wirth does not bite too hard

OK, so then don't.

Doesn’t that make sense though as each manipulates a different layer in the memory hierarchy allowing the programmer to control the latency and throughput implications. I see it as a good thing.

Is this really strange? Matmul is just a specialized kind of primitive compute, one that is seeing an explosion in practical uses.

A Mac Quadra in 1994 probably had floating point compute all over the place, despite the 1984 Mac having none.

Thankfully I think libraries like Pytorch abstract this stuff away. But it seems very convoluted if you're building something from the ground up.

I wonder if some Apple-made software, like Final Cut, make use of all of those "duplicated" instructions at the same time for getting a better performance...

I know how just the multitasking nature of the OS probably make this situation happens across different programs, but nonetheless would be pretty cool!

Would it be possible to use all of them at the same time? Not necessarily in a practical way, but just for fun? Could different ways of doing this on CPU be done in some extent by one core at the same time, given it's superscalar?

Adding CPUs and GPUs on top of your CPUs and GPUs... Sounds like we've the spiritual successor of the Sega Saturn.

I inferred that they meant the neural engine cores by neural accelerators or it could be a bigger/different AMX (which really should become a standard btw)

>Apple also appears to be adding a “Neural Accelerator” to each core on the M5?

The "neural accelerator" is per GPU core, and is matmul. e.g. "Tensor cores".

> the main thing limiting Apple Intelligence (in my opinion) will continue to be the 8 GB of unified memory they still insist on.

As you said - it won’t help previous generations, though since last year (or two??) all macs start with 16GB of memory. Even entry level macbook airs.

Faster compute helps, for things like vision language model that requires bigger context to be filled. My understanding is that ANE is still optimized for convolution load, and compute efficiency while the new neural accelerators optimized for flexibility and performance.

MLX doesn't use the neural engine still right? I still wish they would abandon that unit and just center everything around metal and tensor units on the GPU.

My guess is that they moved the systolic arrays inside the GPU cores just like how it's done in modern NVIDIA chips.

That's the only way to speed up MLX 4x compared to M4.

ofc true. Unified memory is always less than vram. And my 16GB vram aren't enough.

But I think it's also a huge issue Apple makes storage so expensive. If Apple wants local AI to answer your questions it should be able to take your calender, emails, text messages, photos, journal entries etc. into account. It can't do that as nicely as long as customers opt for only 256GB or 1TB devices due to cost

I can only guess that significant changes in hardware have longer lead times than software (for example). I suppose I am not expecting anything game-changing until the M6.

Storage-gating is really disgusting considering how much Apple charges for storage.

Unfortunately, they have also succumbed to the AI hype machine. Apple, calling it by its actual name "machine learning" was about the only thing I still liked about Apple.

I like sniping - but I could make a product call here to support the messaging - when it's running outside diffusion models and LLMs (as per the press release) we could call that AI. Agreed that they should at least have mentioned Apple Intelligence in their PR though

Not all is lost: AI can still be acronym for Apple Intelligence.

I am sure by AI they mean Apple Intelligence:-)

This is not the first time this has happened in Apple’s history. The transition from the 68k architecture to the PowerPC brought major performance improvements, but Apple’s software didn’t take full advantage of it. If I remember correctly, even after the PowerPC switch, core elements of the classic Mac OS still ran in emulation as late as Mac OS 9. Additionally, the classic Mac OS lacked protected memory and preemptive multitasking, leading to relatively frequent crashes. Taligent and Copland were attempts to address these issues, but they both faced development hell, culminating with the purchase of NeXT and the development of Mac OS X. But by the time Mac OS X was released, PowerPC was becoming less competitive than the x86, culminating with the Intel switch in 2006. At this point it was Apple’s software that distinguished Macs from the competition, which remained the case until the M1 Macs were released five years ago.

Software is very easy to bloat, expand scope, and grow to do more than really needed, or just to release apps that are then forgotten about.

Hardware is naturally limited in scope due to manufacturing costs, and doesn't "grow" in the same way. You replace features and components rather than constantly add to them.

Apple needs someone to come in and aggressively cut scope in the software, removing features and products that are not needed. Pair it down to something manageable and sustainable.

Well besides software that runs in data centers/ cloud most other software is turning to crap. And people who think this crap is fine have now reached to position of responsibility at lot of companies. So things would go only worse from here.

I don't think it's the modern Apple, I think that's just Apple.

I remember using iTunes when fixing the name of an album was a modal blocking function that had to write to each and every MP3, one by one, in the slowest write I have ever experienced in updating file metadata. Give me a magnetised needle and a steady hand and I could have done it faster.

A long time ago they had some pretty cool design guides, and the visual design has often been nice, but other than that I don't think their software has been notable for its quality.

Apple has always been a hardware company first - think of how they sell consumers computers with the OS for free, while Microsoft primarily just sells the OS (when comparing the consumer business; I don’t want to get into all the other stuff Microsoft does).

Now that they own the SoC design pipeline, they’re really able to flex these muscles.

What I would do for Snow Leopard on the M class hardware.

I dunno, didnt they already crack the 400GB/s memory bandwidth some years ago? This seems like just another small bump to handle latest OS effects sludge.

Now the M1 range, that really was an impressive 'outperform' moment of engineering for them, but otherwise this is just a clock-work MBA driven trickle of slightly better over-hyped future eWaste.

To outperform during this crisis, hardware engineers worth their salt need to designing long lived boxes with internals that can be easily repaired or upgraded. "yeah but the RAM connections are fiddly" Great, now that sounds like a challenge worth solving.

But you are right about the software. Installing Asahi makes me feel like I own my compter again.

> The modern Apple feels like their hardware teams way outperforming the software teams.

There aren't a lot of tangible gains left to be made by the software teams. The OS is fine, the office suite is fine, the entertainment apps are fine.

If "performance" is shoving AI crap into software that was already doing what I wanted it to do, I'd rather the devs take a vacation.

This seems to be pretty true in general. SBC companies are not competing with Raspberry Pi because their software is quite a bit behind (boot loaders, linux kernel support, etc). Particle released a really cool dev board recently, but the software is lacking. Qualcomm struggled with their new CPU launch with poor support as well. It sometimes takes a while for new Intel processor features to be supported in the toolchains, kernel, and then get used in software.

Aside from that, I think of Apple as a hardware company that must write software to sell their devices, maybe this isn't true anymore but that's how I used to view them. Maintaining and updating as much software as Apple owns is no small task either.

It does feel like Apple is firing on all cylinders for their core competencies.

Software (iOS26), services (Music/Tv/Cloud/Apple Intelligence) and marketing (just keep screaming Apple Intelligence for 3 months and then scream Liquid Glass) ---- on the other hand seem like they are losing steam or very reactive.

No wonder John Ternus is the widely anticipated to replace Tim Cook (and not Craig).

There has to be a whole different mindset with hardware though. Every change has to necessarily be more considered, cross-checked. And I don't say this in any way to disparage software engineers (hold up hand) but I suspect there's a discipline in hardware design that is ... less rigidly adhered to in software design. (And a software update containing a revert, though undesirable, is always a solution.)

Yup. And the marketing department is ahead of both of them.

Apple have always been a hardware company, like Google have always been a software company even if they're doing hardware too now.

There are talks of the hardware head replacing Cook.

Hopefully that will bring whatever they’re doing right to other teams.

For Vision Pro, software team has been impressive. And arguably outperformed the hardware team.

But this is the exception.

It must be observed that the Apple enterprise is, above all else, a purveyor of fine physical contrivances and apparatus.

Furthermore, they do also engage in the traffic and sale of digital programmes wrought by the hands of other, independent artisans.

The hardware team has always shined, but how about one example of this:

The PowerBook from the mid 1990’s were hugely successful, especially the first ones, which were notable for what we now take for granted: pushing the keyboard back allowing space for palm rests. Wikipedia says at one time Apple had captured 40% of the laptop market. All the while the ’90s roared on, Apple was languishing, looking for a modern OS.

Doing good job is rewarded.

Apple's Hardware Chief, John Ternus, seems to be next in line for succession to Tim Cook's position.

I want this hardware available for other systems.

Yes. And their consumer teams are way outperforming their business teams.

Been like that since 1977

From my vast experience with MacOS, Apple is notoriously bad at the most basic software, like notes or calculator

Apple is a hardware company. This has always been the case. It's not just the modern Apple.

In a sense, hardware's job is easier, because the goals are more clear. Make it faster, and more power efficient. Vast amounts of complexity within those goals. But try to summarize the north-star vision for a complex software project like an OS in terms anywhere close as simply as this.

i was about to write exactly that.

I've been thinking whether it could be a reasonable move for Apple to launch a cheaper secondary brand, one that offers devices capable of running Linux or Windows to reach a broader market without cannibalizing its own.

I pretty much see the Macbook as some fancy toys with mediocre software. Maybe the kernel is solid but other software are very meh, even comparing to Windows. But I'm definitely biased as a Windows/Linux user, and my hobby is system programming so naturally a Linux box is more suitable.

Biggest grief with MacOS software:

- Finder is very mediocre comparing to even File explorer in Windows

- Scrollbar and other UI issues

Unfortunately I don't think Asahi is going to catch up, and Macbook is so expensive, so I'll probably keep buying second hand Dell/Lenovo laptop and dump a Linux on top of it.

Apple relies heavily on H1-B slave labor. They don’t pay their software teams enough to be competitive and they run with only about a third of the headcount they need to polish the software. Thus, they have mediocre talent and not enough of it. Penny-wise, pound foolish.

Their software is literally falling apart. ios26 was the biggest trash ive ever experienced from a company this big

it's the same core, so more than likely yes.

M5 is supposed to support FP4 natively which would explain the speed up on Q4 quantized models (down from BF16).

Its not uncommon for Apple and others to compare against two generations ago rather than the immediately preceding one

You wrote:

>Looking at my Macbook though, I can say with utmost certainty that it isn't 4000x faster than the Intel one I had 5 years ago. So, where is the disconnect here?

They wrote:

> Together, they deliver up to 15 percent faster multithreaded performance over M4

The problem is comprehension, not marketing.

The disconnect is that you're reading sideways.

First line on their website:

> M5 delivers over 4x the peak GPU compute performance for AI compared to M4

It's the GPU not the CPU (which you compare with your old Intel) and it's an AI workload, not your regular workload (which again is what you compare)

Apple has also seemingly stopped caring about the quality and efficiency of their software. You can see this especially in the latest iOS/iPadOS/macOS 26 versions of their operating systems. They need their software leadership to match their hardware leadership, otherwise good hardware with bad software still leads to bad product, which is what we are seeing now.

I’m not sure I see the disconnect.

At our company we used to buy everyone MacBook Pros by default.

After the M-series chip, the MBPs are just too powerful and no longer necessary for the average white collar worker (they seem like “actual” pro machines, now) to the point where we now order MacBook Airs for new employees.

I feel like until recently, you really needed a MBP to get a decent UX (even just using chrome). But now there doesn’t seem to be a major compromise when buying an Air for half the price, at least compared to 3-5 years ago.

It's GPU performance.

Spin up ollama and run some inference on your 5-year-old intel macbook. You won't see 4000x performance improvement (because performance is bottlenecked outside of the GPU), but you might be in the right order of magnitude.

> So, where is the disconnect here?

> I can say with utmost certainty that it isn't 4000x faster

The numbers you provided do not come to 4000x faster (closer to 2400x)

> Why is actual user experience not able to keep up with benchmarks and marketing?

Benchmarks and marketing are very different things, but you seem to be holding them up as similar here.

The 5x 6x 4x numbers you describe across marketing across many years don't even refer to the same thing. You're giving numbers with no context, which implies you're mixing them and the marketing worked because the only thing you're recalling is the big number.

Often, every M-series chip is a HUGE advancement over the past in GPU. Most of the "5x" performance jumps you describe are in graphics processing, and the "Intel" they're comparing it to is often an Intel iGPU like the Iris Xe or UHD series. These were low end trash iGPUs even when Apple launched those Intel devices, so being impressed by 5x performance when the M1 came out was in part because the Intel Macs had such terrible integrated graphics.

The M1 was a giant jump in overall system responsiveness, and the M-series seems to be averaging about a 20% year over year meaningful speed increase. If you use AI/ML/GPU, the M-series yearly upgrade is even better. Otherwise, for most things it's a nice and noticeable bump but not a Intel-to-M1 jump even from M1-to-M4.

All those extra flops are spent computing light refraction in the liquid glass of the ui

> M5 is 4-6x more powerful than M4

In GPU performance (probably measured on a specific set of tasks).

Those marketing claims are each about a very specific workload, not about general performance. Yes, it is often misleading.

Well, if you read the very next thing after 4x, you will notice it says "the peak GPU compute performance for AI compared to M4".

The disconnect here is that you can't read. Sorry, no other way to say it.

Probably synthetic benchmarks that don't represent actual bottlenecks in application usage. How much of what you are doing is actually CPU bound? Your machine still has to do I/O, and even though that's "very fast" these days, it's not happening inside your CPU, so you'll only see the actual improvements when running workloads that benefit from the performance improvements (i.e., complex calculations that can live in the CPU and its cache).

What scares me is that my M2 started seeing performance issues in macOS recently. Safari is sometimes slow (I admit I stress it with many tabs, but it wasn't like this a year ago.) Somehow the graphics in general seems slower on Tahoe, eg the effects when minimising a window.

I am deeply concerned all the performance benefits of the new chips will get eaten away.

You know, 64% of statistics are made up.

Says M5 is 4x faster than M4 and 6x faster than M1 for AI compute on the GPU. Basically M4 was only a little faster than M1 at this task. Ex. if M5 is 24 AI TOPS, M4 is 6 AI TOPS, and M1 is 4 AI TOPS.

Unless you're looking at your MacBook running LM Studio you won't be seeing much improvement in this regard.

Each is a different specific benchmark, so they don't stack the way you're doing.

This is 4-6x faster in AI for instance.

User experience (for most things, unless you sit there encoding video all day) isn't really related to raw performance so much as latency. Processor power can help there, but design and at the limit memory latency is the key constraint.

It states it is "peak performance". Probably in a very specific use case. Or maybe it reaches the peak for an extremely short period of time before it drops the performance.

They are not 4x more powerful than the previous generation at everything, or even at the same thing every time, so it does not stuck up. Here 4x refers sth wrt LLMs running on the GPU.

I use both an M1 max and an M3 max, and frankly I do not notice much difference if you control for the core count in most stuff. And for running LLMs they are almost the same performance. I think from M1-M3 there was no much performance increase in general.

Agreed, if I have 40 tabs opened on Chrome, my M1 macbook is no longer responsive... I'm not sure about their performance claims, apart from some niche GPU rendering for games, which constitutes about 0% of my daily laptop usage

Because this is bullshit, lies, marketing

Because there's more to "actual user experience" than peak CPU/GPU/NPU workload.

Firstly, the M5 isn't 4-6x more powerful than M4 - the claim is only for GPU, only for one narrow workload, not overall performance uplift. Overall performance uplift looks like ~20% over M4, and probably +100% over M1 or so.

But there is absolutely a massive sea change in the MacBook since Intel 5 years ago: your peak workloads haven't changed much, but the hardware improvements give you radically different UX.

For one thing, the Intel laptops absolutely burned through the battery. Five years ago the notion of the all-day laptop was a fantasy. Even relatively light users were tethered to chargers most of the day. This is now almost fully a thing of the past. Unless your workloads are very heavy, it is now safe to charge the laptop once a day. I can go many hours in my workday without charging. I can go through a long flight without any battery anxiety. This is a massive change in how people use laptops.

Secondly is heat and comfort. The Intel Macs spun their fans up at even mild workloads, creating noise and heat - they were often very uncomfortably warm. Similar workloads are now completely silent with the device barely getting warmer than ambient temp.

Thirdly is allowing more advanced uses on lower-spec and less expensive machines. For example, the notion of rendering and editing video on a Intel MacBook Air was a total pipe dream. Now a base spec MacBook Air can do... a lot that once forced you into a much higher price point/size/weight.

A lot of these HN conversations feel like sports car fans complaining: "all this R&D and why doesn't my car go 500mph yet?" - there are other dimensions being optimized for!

Being late in AI race or not entering it from training side is not necessarily bad, others have burned tons of money, if Apple enters with their hardware first (only?) it may disrupt status quo from consumer side. It's not impossible that they'll produce hardware everybody will want to run local models that will be on par with closed ones. If this happens it may change real money flow (as opposed to investor based on imaginary evaluation money that can evaporate).

They are the leader in manufacturing consumer systems with sufficient high-bandwidth memory to enable decent-sized LLMs to be run locally with reasonable performance. If you want to run something that needs >=32GB of memory (which is frankly bottom-end for a somewhat capable LLM) they're your only widely-available choice (otherwise you've got the rare Strix Halo AI Max+ 395 chip, or you need multiple GPUs, or maybe a self-build based around a Threadripper.)

This might not be widely recognised, as the proportion of people wanting to run capable LLMs locally is likely a rounding error versus the people who use ChatGPT/Claude/Gemini regularly. It's also not something that Apple market on, as they can't monetize it. However, as time goes on and memory and compute power gradually decrease in price, and also maybe as local LLMs continue to increase in ability (?) it may become more and more relevant.

153 GB/s is not bad at all for a base model; the Nvidia DGX Spark has only 273 GB/s memory bandwidth despite being billed as a desktop "AI supercomputer".

Models like Qwen 3 30B-A3B and GPT-OSS 20B, both quite decent, should be able to run at 30+ tokens/sec at typical (4-bit) quantizations.

Looks like the M5 base has LPDDR5x-9600, which works out to 153.6 from base M4's 120GB/s DDR5x-7500. The Pro/Max versions have more memory controllers, 16, 24 and 32 channels accordingly. The 32 channel M5 top-end version will have 614GB/s by my calculations.

It would take 48 channels of DDR5x-9600 to match a 3090's memory bandwidth, so the situation is unlikely to change for a couple of years when DDR6 arrives I guess

The memory capacity to me is an even bigger problem, at 32GB max.

I am interested to learn why models move so much data per second. Where could I learn more that is not a ChatGPT session?

You don’t want to be bandwidth-bound, sure. But it all depends on how much compute power you have to begin with. 153GB/s is probably not enough bandwidth for an Rtx5090. But for the entry laptop/tablet chip M5? It’s likely plenty.

My guess would be those are going into the rumoured OLED models coming out next year.

Enough or not, they do describe it like this in an image caption:

"M5 is Apple’s next-generation system on a chip built for AI, resulting in a faster, more efficient, and more capable chip for the 14-inch MacBook Pro, iPad Pro, and Apple Vision Pro."

With MoE LLMs like Qwen 3 30B-A3B that's no longer true.

What do you mean by properly? What’s the behavior one would observe if they did run an llm?

Most of the AI and Machine Learning Apple has done so far are primarily done on device so you can see whether there is any climate change concern or not.

Assuming you've read https://andymasley.substack.com/p/a-cheat-sheet-for-conversa... or the longer full essay/related works, could you elaborate on why you don't use Apple Intelligence?

I totally understand why someone would refuse to use it due to environmental reasons (amongst others) but I'm curious to hear your opinions on it.

> making Apple AI [...] their focus

Are they really doing that? Because if it's the case they have shockingly little to show for it.

Their last few attempts at actual innovation seem to have been less than successful. The Vision Pro failed to find a public. Liquid Glass is to put it politely divisive.

At that point to me, it seems that good SoC and a captive audience in the US are pretty much all they have remaining and competition on the SoC part is becoming fierce.

If you don’t use AI for climate reasons then you should read the recent reports about how little electricity and water is actually used. It’s basically zero (image and video models excluded). Your information about this is probably related to GPT3.5 or something. Which is now 3 years old - a lifetime in AI world.

I think they will continue ruining their products via software updates. That's implied by a walled garden approach they chose to do their business: this forces users to consoom more and thus generates profits. Apple isn't a "lean" company, it needs outrageous profits to stay afloat.

I'm interested in reading about your low-carbon lifestyle that is so efficient you got to the point of giving up machine inference.

Yeah, this is the biggest hole in ARM offerings.

The only well supported devices are either phones or servers with very little in between.

Even common consumer devices like wifi routers will have ARM SOCs with pinned version of the kernel they are attached to which will get supported for 1 to 2 years at most.

I have a pretty good time on Asahi Fedora (macbook air M1). It supposedly also supports M2 but no higher.

And it's a PITA to install (needs to be started within macosx, using scripts, with the partitions already in a good state)

Apparently the Windows exclusivity period has ended, so Google will support Android and ChromeOS on Qualcomm X2-based devices, https://news.ycombinator.com/item?id=45368167

Off the top of my head, here is what that needs:

  1. Implementing PR_SET_SYSCALL_USER_DISPATCH
  2. Implementing ntsync
  3. Implementing OpenGL 4.6 support (currently only OpenGL 4.1 is supported)
  4. Implementing Vulkan 1.4 with various extensions used by DXVK and vkd3d-proton.

  1. Not implementing this would just break games like Jurassic World where DRM hard codes Windows syscalls. I do not believe that there are many of these, although I could be wrong.
  2. There is https://github.com/marzent/wine-msync, although implementing ntsync in the XNU kernel would be better.
  3. The latest OpenGL isn't that important these days now that Vulkan has been widely adopted, although having the latest version would be nice to have for parity. Not many things would suffer if it were omitted.
  4. They could add the things needed for MoltenVK to support Vulkan 1.4 with those extensions on top of Metal:

It is a shame that they do not work with Valve on these things. If they did, Proton likely would be supported for MacOS from within Steam and the GPTK would benefit.

> lack of anticheat support.

I just redid my windows machine to get at TPM2.0 and secure boot for Battlefield 6. I did use massgrave this time because I've definitely paid enough Microsoft taxes over the last decade. I thought I would hate this new stuff but it runs much better than the old CSM bios mode.

Anything not protected by kernel level anti cheats I play on my steam deck now. Proton is incredible. I am shocked that games like Elden Ring run this well on a linux handheld.

Good point. Many people (including me) switched to Apple Silicon with the hope (or promise?) of having just one computer for work and leisure, given the potential of the new architecture. That didn't happen, or only partially, which is the same.

In my case, for software development, I'd be happy with an entry-level MacBook Air (now with a minimum of 16GB) for $999.

I can't sell my MacBook Pro because the speakers are so insanely good. Air can't compare. The speakers are worth the extra kilos.

I agree—the difference between the different compatibility layers and native games is very steep at times. Death Stranding on my M2 Pro looks so good it’s hard to believe, but running GTA Online is so brittle and clunky… Even when games have native macOS builds, it’s rare to find them with Apple Silicon support (and even rarer with Metal support). There is a notable exception though: Arma 3 has experimental Apple Silicon support, though it comes with significant limitations. (Multiplayer, flying & mods) Although I don’t believe it’s in Apple’s interest, gaming on Linux might become an option in the future, even on Mac, but the lack of ARM builds is an even bigger problem there…

Since I am playing mostly MSFS 2024 these days I currently use GeForce Now which is fine, but cloud gaming isn’t still quite there yet…

On top of that, what is the strategy from Apple on gaming? Advertise extra performance and features that you only get if you upgrade your whole device? This is non-sustainable to put it mildly. There are egpu enclosures with TB5, developing something like that for the Mac would make more sense if they really cared about gaming anyhow.

I'm gonna be looking for a 4080 in SFF form factor since my current gaming rig can't get upgraded to win 11. Also I wouldn't mind a smaller desktop.

edit: for now I'll get that win 10 ESU

Honestly, gaming consoles are so much cheaper and "no hassle." I never games on my Mac.

What about wine flavor from crossdressers?

Yeah I agree. If it weren't for gaming I would have already uninstalled Windows permanently. It's really unfortunate because it sticks out as the one product in my house that I truly despise but I can't get rid of, due to gaming.

I've been trying to get Unreal Engine to work on my Macbook but Unity is an order of magnitude easier to run. So I'm also stuck doing game development on my PC. The Metal APIs exist and apparently they're quite good... it's a shame that more engines don't support it.

Going back to the Air's screen from your Pro will be a steep fall.

Sometimes I just feel like buying the latest and greatest game, I have an m4 too, the choices are usually quite abysmal. I agree.

> I wish Apple would take gaming more seriously and make GPTK a first class citizen such as Proton on Linux.

Note that games with anticheat don't work on Linux with Proton either. Everything else does, though.

Many people blame the lack of OpenGL/Vulkan... but I really don't buy it. It doesn't pass the sniff test as an objection. PlayStation doesn't support OpenGL/Vulkan (they have their own proprietary APIs, GNM, GNMX, PSSL). Nintendo supports Vulkan but performance is so bad, almost everyone uses the proprietary API (NVN / NVN2). Xbox obviously doesn't accept OpenGL/Vulkan either, requiring DirectX. Understanding of Metal is widespread in mobile gaming, so it's weird AAA couldn't pull from that industry if they wished.

Yep, I use Moonlight / Sunshine / Apollo to stream from my gaming PC, so I still use my Mac setup but get nearly perfect windows gaming with PC elsewhere in house.

This has been by far the best setup until Apple can take gaming seriously, which may never happen.

The main roadblock for porting the games to Mac has never been the hardware, but Apple themselves. Their entire attitude is that they can do whatever they please with their platforms, and expect the developers to adjust to the changes, no matter how breaking. It’s a constant support treadmill, fixing the stuff that Apple broke in your previously perfectly functioning product after every update. If said fixing is even possible, like when Apple removed support for 32-bit binaries altogether, rendering 3/4 of macOS Steam libraries non-functional. This works for apps, but it‘s completely antithetical to the way game development processes on any other platform are structured. You finish a project, release it, do a patch cycle, and move on.

And that’s not even talking about porting the game to either Metal or an absolutely ancient OpenGL version that could be removed with any upcoming OS version. A significant effort just to address a tiny market.

As far as I’ve seen, Apple is to blame here as they usually make it harder to target their platform and don’t really try to cooperate with the rest of the industry.

As a game developer, I have to literally purchase Apple hardware to test rather than being able to conveniently download a VM

Mac dev sucks. You're forced to use macos and xcode (for the final build anyway). You're not able to virtualize the build machines.

Apple is actively hostile to how you would build for Linux or PC or console.

I was very surprised, and pleasantly too, that Cyberpunk 2077 can maintain 60FPS (14", M4 Pro, 24gb RAM) with only occasional dips. Not with full resolution (actually around FullHD), but at least without "frame generation". Turning frame generation on, it now can output 90-100 FPS depending on environment, but VSync is disabled so dips become way more noticeable.

It even has "for this mac" preset which is good enough that you don't need to tinker with settings to have decent experience.

The game is paused, almost like becomes "frozen" if it's not visible on screen which helps with battery (it can be in the background without any noticeable impact on battery and temperature). Overall way better experience than I expected.

Multiple solid reasons have been mentioned from ones created by Apple to ones enforced in software by Apple. One that hasn't been mentioned is the lack of marketshare. Macos market is just tiny and very limited. It's also not a growing market. PC gaming isn't blowing up either but the amount of players is just simply higher.

Ports to macos have not done well from what I've heard. However you can see ports on PC do really well and have encouraged studios like Sony and SquareEnix to invest more in PC ports. Even much later after the console versions sell well. Just not a lot of reasons to add the tech debt and complexity of supporting mac as well.

Even big publishers like Blizzard who have been mac devs for a long time axed the dedicate mac team and client and moved to a unified client. This has downfalls like mac specific issues. If those are not critical then they get put in the pile with the rest of the bugs.

It's easier to develop a game for a mac in some ways but you reach a tiny fraction of gamers that way.

I play a lot of World of Warcraft on my M3 MacBook Pro which has a native MacOS build. It's a CPU bottlenecked game with most users recommending the AMD X3D CPUs to achieve decent framerates in high end content. I'm able to run said content at high (7/10) graphics settings at 120fps with no audible fan noise for hours at a time on battery. It's been night and day compared to previous Windows machines.

There's a cost/value calculation that just doesn't work well...I have a Ryzen9/rtx3070 PC ($2k over time) and my M4 Mini ($450) holds it's own for most all normal user stuff...sprinting ahead for specific tasks (Video CODEC)...but the 6 year old dedicated GPU on the PC annihilates the Mini in pushing pixels...You can spec an Apple that does better for gaming, but man, are you gonna pay for it, and still not keep up with current PC GPUS.

Now...something like minecraft or SubNautica? The M4 is fine, especially if you're not pushing 4k 240hz.

Apple has been pushing the gaming experience for years (iPhone 4s?) but it never REALLY seems to land, and when someone has a great gaming seperience in a modern AAA game, they always seem to be using a $4500 Studio or similar.

I wrote a post (rant)[1] about my experience of releasing a game on macOS as an indie dev. tl;dr: Apples goes a long way to make the process as painful as possible with tons of paper cuts.

[1] https://ruoyusun.com/2023/10/12/one-game-six-platforms.html#...

Metal is a very recent API compared to DirectX and OpenGL. Also, there’s very very little people on Mac, and even less that also play videogames. There are almost no libraries and tooling built around Metal and the Mac SDKs, and a very small audience, so it doesn’t make financial sense.

you have to release major titles for windows and console, because there are tons of customers using them.

so a mac port, even if simple, is additional cost. there you have the classic chicken and egg problem. the cost doesn't seem to be justified by the number of potential sales, so major studios ignore the platform. and as long as they do, gamers ignore the platform

i've seen it suggested that Apple could solve this standoff by funding the ports, maybe they have done this a few times. but Apple doesn't seem to care much about it

Up to some years ago, it was common for gamers to assemble their own PC, something that you can't do with a Mac. Not sure if this is still common among gamers though.

The porting is not straightforward; you must switch to Metal, you should adapt rendering pipeline to tiled deferred shading.

- have to build using XCode on macOS

- have to pay Apple to have your executable signed

- poor Vulkan support

The hardware has never been an issue, it's Apple's walled garden ecosystem.

i think it depends on how easy it is for a dev to deploy to apple. M1 was great at running call of duty in a windows emulator. iPhone can run the newest resident evil. apple needs to do more to convince developers to deploy to mac

Doesn’t MacOS favor an 60Hz output? Gamers prefer much higher rates.

And don’t forget they made an VR headset without controllers.

Apple doesn’t care about games

It's kind of a myth though, Mac has many flagship games and everything in between

If you identify as a "gamer" and are in those communities, then you'll see communities talking about things you can't natively play

but if you leave niches you already have everything

and with microtransactions, Apple ecosystem users are the whales. again, not something that people who identify as "gamers" wants to admit being actually okay with, but those people are not the revenue of game production.

so I would say it is a missed opportunity for developers that are operating on antiquated calculations of MacOS deployment

Meanwhile intel seems to be doing a big bunch of nothing much. And AMD seems busy playing house with OpenAI to catch up to nvidia on the GPU front.

Now if only Apple would sell these for use outside of their walled garden.

Because gaming on Mac actually looks bad in practice.

https://news.ycombinator.com/item?id=44906305

They want you to buy the Apple phone and pair it, so they sell more

Because the A* iPhone chip in the Apple TV should be more than enough for HD quality gaming?

I think it is because there are not enough games to be the reason for integrating an M* chip.

Yeah, they rebranded it "Apple Intelligence" but this press release appears to be mostly using AI in the same (vague) way that the rest of the industry does.

Also just noticed this:

"And now with M5, the new 14-inch MacBook Pro and iPad Pro benefit from dramatically accelerated processing for AI-driven workflows, such as running diffusion models in apps like Draw Things, or running large language models locally using platforms like webAI."

First time I've ever heard of webAI - I wonder how they got themselves that mention?

I'd argue that Apple's definition of "neural engine" was entirely different from what the greater desktop, edge and datacenter markets already considered a "neural engine" to be.

It's an improvement, nomenclature-wise.

That's laughable in 2025, and together with the wimpy 153 GB/s memory bandwidth (come on, Strix Halo is 256GB/s at a fraction of the price!) they really don't have a leg to stand on calling this AI-anything!

why do you want more CPU cores and better performance than the M4, if not for running local AI models?