/s I assume but it’s crazy to me that LG runs on the watch

Unlikely.

It's not exactly a bump if they raise prices at the same time, though with the RAM situation I'm not mad.

So they renamed performance to mean efficiency and are now using super in place of performance?

I think super cores are a new type/tier of core, not a rename of performance.

The base M5 has super/efficiency cores.

The Pro and Max have super/performance cores.

I don't think the "new" Performance cores are just "renamed" "E" / "Efficiency" cores; Apple has retroactively renamed the baseline M5 nomenclature to say it has "10-core CPU with 4 super cores and 6 efficiency cores"; so they're clearly keeping the "efficiency cores" nomenclature around.

I think this is a new design, with Apple having three tiers of cores now, similar to what Qualcomm has been doing for a while.

I think how it breaks down is:

- "Super" are the old "P" cores, and the top tier cores now

- "Performance" cores are a new tier and seen for the first time here, slotting between "old" P and E in performance

- "Efficiency" / "E" are still going to be around; but maybe not in desktop/Pro/Max anymore.

I was looking into this. The M5 performance cores can be scaled down to match efficiency cores in performance and power usage.

I believe they lower the clock speed, limit how much work is done in parallel on each core, and limit how aggressive the speculative execution is so less work is wasted.

looks like this will be their angle for the whole agentic AI topic

"Apple Intelligence is even more capable while protecting users’ privacy at every step."

Remains to be seen how capable it actually is. But they're certainly trying to sell the privacy aspect.

We had a workshop 6 months ago and while I've always been sceptical of OpenAI,etc's silly AGI/ASI claims, the investments have shown the way to a lot of new technology and has opened up a genie that won't be put back into the bottle.

Now extrapolating in line with how Sun servers around year 2000 cost a fortune and can be emulated by a 5$ VPS today, Apple is seeing that they can maybe grab the local LLM workloads if they act now with their integrated chip development.

But to grab that, they need developers to rely less on CUDA via Python or have other proper hardware support for those environments, and that won't happen without the hardware being there first and the machines being able to be built with enough memory (refreshing to see Apple support 128gb even if it'll probably bleed you dry).

  Are they doubling down on local LLMs then?

Neural Accelerator was present in iPhone 17 and M5 chip already. This is not new for M5 Pro/Max.

Apple's stated AI strategy is local where it can and cloud where it needs. So "doubling down"? Probably not. But it fits in their strategy.

I think its just marketing, and the marketing is working. Look how many people bought Minis and ended up just paying for API calls anyway. (Saw it IRL 2x, see it on reddit openclaw daily)

I don't mind it, I open Apple stock. But I'm def not buying into their rebranding of integrated GPU under the guise of Unified Memory.

The topic is MacBook, so my criticism is a little off. However, I really dont believe in this "local LLM" promise from Apple. My phone already gets noticeably warm if I answer 5 WhatsApp messages. And looses 5% of battery during the process. I highly doubt Apple will have a useable local LLM that doesn't drain my battery in minutes, before 2030.

The hardware capabilities that make local LLMs fast are useful for a lot of different AI workloads. Local LLMs are a hot topic right now so that’s what the marketing team is using as an example to make it relatable.

I've been so disappointed in Apple's lack of execution on this. There is so much potential for fantastic local models to run and intelligently connect to cloud models.

I just don't get why they're dropping the ball so much on this.

> Are they doubling down on local LLMs then?

Honestly, I think that's the move for apple. They do not seem to have any interest in creating a frontier lab/model -- why would they give the capex and how far behind they are.

But open source models (Kimi, Deepseek, Qwen) are getting better and better, and apple makes excellent hardware for local LLMs. How appealing would it be to have your own LLM that knows all your secrets and doesnt serve you ads/slop, versus OpenAI and SCam Altman having all your secrets? I would seriously consider it even if the performance was not quite there. And no need for subscription + cli tool.

I think apple is in the best position to have native AI, versus the competition which end up being edge nodes for the big 4 frontier labs.

It is simply marketing nonsense - what they really mean (I think) is they support matrix multiplication (matmul) at the hardware level which given AI is mostly matrix multiplications you'll get much faster inference (and some increase in training too) on this new hardware. I'm looking forward to seeing how fast a local 96gb+ LLM is on the M5 Max with 128gb of RAM.

But memory bandwidth (bottleneck for LLM inference) is only marginally improved, 614 GB/s vs 546 GB/s for M4/M5 Max - where is this 4x improvement coming from?

I think I'll pass on upgrading.

It’s not necessarily doubling down on local. The reality is your LLM should be inferencing every tick … the same way your brain thinks every. Fucking. Nano. Second.

So yes, the LLM should be inferencing on your prompt, but it should also be inferencing on 25,000 other things … in parallel.

Those are the compute needs.

We just need compute everywhere as fast as possible.

There already are a bunch of task-specific models running on their devices, it makes sense to maintain and build capacity in that area.

I assume they have a moderate bet on on-device SLMs in addition to other ML models, but not much planned for LLMs, which at that scale, might be good as generalists but very poor at guaranteeing success for each specific minute tasks you want done.

In short: 8gb to store tens of very small and fast purpose-specific models is much better than a single 8gb LLM trying to do everything.

Apple's AI strategy really kind of threads the needle cleverly.

"AI" (LLMs) may or may not have a bubble-pop moment, but until it does Apple get to ride it on these press releases and claims. But if the big-pop occurs, then Apple winds up with really fantastic hardware that just happens to be good at AI workloads (as well as general computing).

For example, image classification (e.g. face recognition/photo tagging), ASR+vocoders, image enhancement, OCR, et al, were popular before the current boom, and will likely remain popular after. Even if LLM usage dries up/falls out of vogue, this hardware still offers a significant user benefit.

Apple absolutely has a massive opportunity here because they used a shared memory architecture.

So as most people in or adjacent to the AI space know, NVidia gatekeeps their best GPUs with the most memory by making them eye-wateringly expensive. It's a form of market segmentation. So consumer GPUs top out at 16GB (5090 currently) while the best AI GPUs (H200?) is 141GB (I just had to search)? I think the previou sgen was 80GB.

But these GPUs are north of $30k.

Now the Mac Studio tops out currently at 512GB os SHARED memory. That means you can potentially run a much larger model locally without distributing it across machines. Currently that retails at $9500 but that's relatively cheap, in comparison.

But, as it stands now, the best Apple chips have significantly lower memory bandwidth than NVidia GPUs and that really impacts tokens/second.

So I've been waiting to see if Apple will realize this and address it in the next generation of Mac Studios (and, to a lesser extend, Macbook Pros). The H200 seems to be 4.8TB/s. IIRC the 5090 is ~1.8TB/s. The best Apple is (IIRC) 819GB/s on the M3 Ultra.

Apple could really make a dent in NVidia's monopoly here if they address some of these technical limitations.

So I just checked the memory bandwidth of these new chips and it seems like the M5 is 153GB/s, M5 Pro is ~300 and M5 Max is ~600. I was hoping for higher. This isn't a big jump from the M4 generation. I suspect the new Studios will probably barely break 1TB/s. I had been hoping for higher.

What % of users actually care that much about local LLMs? It appears to still be an inferior (though maybe decent) service compared to ChatGPT etc., and requires very top-end hardware. Is privacy _that_ important to people when their Google search history has been a gateway to the soul for years? I wonder if these machines would cost significantly less (or put the cost to other things, e.g. more CPU cores) without this emphasis on LLMs.

A useful llm that needs 64gb of ram and mid double digit cores is not useful for 99% of their customers. The LLMs they have on iphone 17's certainly cannot do anything useful other than summerization and stuff. It's a hardware constraint that they have.

> Are they doubling down on local LLMs then?

Apple is in the hardware business.

They want you to buy their hardware.

People using Cloud for compute is essentially competitive to their core business.

> I still think Apple has a huge opportunity in privacy first LLMs

This correlation of Apple and privacy needs to rest. They have consistently proven to be otherwise - despite heavily marketing themselves as "privacy-first"

https://www.theguardian.com/technology/2019/jul/26/apple-con...

Didn't they announce a partnership with Google Gemini?

> doubling down on local LLMs

Do think it'll be common to see pros purchasing expensive PCs approaching £25k or more if they could run SoTA multi-modal LLMs faster & locally.

> Are they doubling down on local LLMs then?

Neural Accelerators (aka NAX) accelerates matmults with tile sizes >= 32. From a very high level perspective, LLM inference has two phases: (chunked) prefill and decode. The former is matmults (GEMM) and the latter is matrix vector mults (GEMV). Neural Accelerators make the former (prefill) faster and have no impact on the latter.

At today's prices, the memory will probably cost more than the rest of the hardware combined :P

128gb was there for a while. I am kind of disappointed they do not have 256gb option.

Checking Apple's store, I can't find a cheaper configuration than $5100 for the M5 + 128GiB version.

Here in Europe, including 21% VAT, that's €6.124,00 ($7.094,35 equivalent).

Because of pricing strategies and such, the 128GiB version comes with a 2TiB SSD at minimum, and also requires the M5 Max (not Pro) at its highest configuration.

Not sure if this is new, but it should be noted that these laptops don't come with a charger any more.

    70W USB-C Power Adapter (included with M5 Pro with 16-core GPU)

    96W USB-C Power Adapter (included with M5 Pro with 20-core GPU, configurable with M5 Pro with 16-core GPU)

    USB-C to MagSafe 3 Cable (2 m)

In US, going to 128 GB from 32 is $1500 extra. However 32 GB is only offered with the 32 core version and 128 only with the 40 core version.

I've upgraded to Tahoe at 26.2, zero complaints from my side. Haven't had any runaway memory leaks or similar that were reported.

Just yesterday, my colleague's mac Time Machine couldn't recover backup and they had to reinstall everything.

But I think this predates Tahoe.

This. I have been a big (and loud) fan of M-series hardware from the beginning, but if Apple is going to keep making their software worse, I will find myself lingering on older generations that run Asahi Linux or going back to a traditional x86_64 laptop instead of buying into new generations.

Yeah this is a real issue with these new Macs. I would wait until macOS 27 to see the direction Apple takes.

The next macOS will be touch screen centric with elements getting bigger when you're close to touching them, rumors say. That being said, I run Tahoe and it works perfectly fine to me, I am not sure what issues people have with it. Sure, some corner radii aren't exactly the same but I honestly couldn't give less of a shit as long as it runs the programs I need.

Increasing component prices perhaps? Get some sales in before you have to jack up the sale price.

You remember well, they didn’t update these last fall.

And another rumor said these are going to be updated again this fall but I’m not sure about that. With OLED screens and M6 (supposedly).

Maybe they want people to have more money available for the new phones later this year, since that market is in decline.

They didn't update them last October is why.

I think at this point Apple will just release new versions of laptops whenever new CPU revisions and yields allow. M5 Pro wasn't ready for October so delayed until now.

M6 is rumored to be released in Q4.

>Anyway, it starts with 16GB of RAM. $400 to go from 16GB to 32GB

Interesting that this hasn't budged since the memory shortages appeared.

I know RAM is scarce and everything, but doubling down on LLM local acceleration with all of that dedicated silicon while at the same time sticking with Apple's traditional lack of RAM availability makes for a very weird product proposition to me.

It starts at 16GB for the base M5 and 24GB for the Pro/Max. It's been like this.

on Silicon Mac's it's never called RAM, it "unified memory"

Apple's RAM price bumps were already insane, now they'll get worse.

> It’s $400 to go from 16GB to 32GB.

No change from the previous models then, 16GB->32GB was already $400. They're cutting into their previously enormous margins to keep the prices stable, rather than hiking the prices to maintain their margins.

> M5 Pro supports up to 64GB of unified memory with up to 307GB/s of memory bandwidth, while M5 Max supports up to 128GB of unified memory with up to 614GB/s of memory bandwidth

Isn't this it?

On the M5 Pro tier (not the base M5 tier that was released last November), the base memory is 24GB.

My M3 Pro from a few years ago for the same price had 18GB.

Insane for the "Pro" to have only 16GB of memory. My 11 year old Intel i3 laptop has 16GB of memory.

Apple doesn't tend to use "RAM" in their marketing materials, they usually use "memory", which appears 9 times in the press release.

Preorders open tomorrow according to the store page. You can’t order the base RAM model today, either.

marketing.

Image Playground

Local LLMs. Lots of people buy Macs due to their unified memory which obviates the need to buy a much more expensive GPU to get the same amount of VRAM.

If you’re working on something sensitive, you may not want to share it with OpenAI or Anthropic.

You can run open source models like Kimi K or Qwen locally. Apple recently updated Xcode 26.3 to support local models.

Yeah I have an M1 Max, and I really want to upgrade, but there’s no reason to.

AI video generation can fairly easily choke anything that's not NVIDIA's flagship model. Even the latest local image gen models are so large that they can be frustratingly slow with non-optimal hardware even if they fit in the VRAM. IIRC when I had an M2, it was about 4x slower at running the venerable Stable Diffusion (and SDXL) than my meager RTX 3060.

You might have confused Hacker News with your e-mail inbox again. This is an Apple press release, directed to everybody in the world who might be interested in a new computer or their first computer.

I've found current-generation Macs so capable that I've switched to using a Macbook Air. Would strongly recommend - it's still a powerful machine and it's significantly lighter and cheaper.

   and that’s with doing pretty demanding 3D and LLM work.

It definitely chokes with larger models that can fit the 192GB of RAM. Prompt processing is a big bottleneck before M5.

If there’s anything this past three years has taught me, it’s that modern cpus can performantly do every task except for streaming text over the internet.

I have a powerful older Mac that doesn’t really “choke” on anything, but I could always use more speed.

The high memory Macs have been great for being able to run LLMs, but the prompt processing has always been on the slow side. The new AI acceleration in these should help with that.

There are also workloads like compiling code where I’ll take all the extra speed I can get. Every little bit of reduced cycle time helps me finish earlier in the day.

And then there’s gaming. I don’t game much, but the M1 and M2 era Apple Silicon feels sluggish relative to what I have on the nVidia side.

Sounds pretty beefy. What kind of local LLM is that thing capable of running? Does it open up real alternatives to cloud providers like OpenAI and Claude, or are the local models this hardware is capable of running still pretty far behind?

>LLM work.

Doubt

I imagine you basically use online models exclusively, and occasionally try out local stuff.

Source: My fortune 20 company tried with M whatever, and the local llms were unusable.

Contrary to HN popular belief, there are neither incentives nor benefits to building native ui apps, for neither consumer nor professional apps. The exception is apps that only make sense on a single platform, such as window management and other deep integration. On iOS/macos you have a segment of indie/smaller apps that capture a niche market of powerusers for things like productivity apps. But the point is it makes no sense for anything from Slack, VSCode, Maya, DaVinci Resolve, and so on, to build native UIs. Even if they wanted to build and maintained 3 versions, advanced features aren’t always available in these frameworks. In the case of Windows, even MS has given up on their own tech, and have opted to launch webview based apps. Apple is slightly more principled.

Similar thoughts with first image of Capture One, when apple bought Pixelmator/Photomator a year ago.

I think I read somewhere long time ago that Capture One is also using Qt for GUI, though cannot find this anymore, so probably not true.

>On M4 Max 128GB we're seeing ~100 tok/s generation

You mean on your first token. Whats the performance after 500 and 3000 tokens?

I genuinely don't understand why people post stuff like this. People are not informed enough to know you mean first tokens. They are going to make a mistake and buy one thinking they will get 100tk/s.

Are you working for Apple marketing? Do you have post purchase regret? I cannot imagine deliberately misleading people. Maybe you are hoping more buyers build up your ecosystem?

I find time to first token more important then tok/s generally as these models wait an ungodly amount of time before streaming results. It looks like the claims are true based on M5: https://www.macstories.net/stories/ipad-pro-m5-neural-benchm... so this might work great.

What about real workloads? Because as context gets larger, these local LLMs aproxiate the useless end of the spectrum with regards to t/s.

4x faster is about token prefill, i.e. the time to first token. It should be on par with DGX Spark there while being slightly faster than M4 for token generation. I.e. when you have long context, you don't need to wait 15 minutes, only 4 minutes.

The memory bandwith on M4 Max is 546 GB/s, M5 Max is 614GB/s, so not a huge jump.

The new tensor cores, sorry, "Neural Accelerator" only really help with prompt preprocessing aka prefill, and not with token generation. Token generation is memory bound.

Hopefully the Ultra version (if it exists) has a bigger jump in memory bandwidth and maximum RAM.

100 tok/s sounds pretty good. What do you get with 70B? With 128GB, you need quantization to fit 70B model, right?

Wondering if local LLM (for coding) is a realistic option, otherwise I wouldn't have to max out the RAM.

Because it was always a vapid distraction from life.

For me going way back, it was exciting when I had to save a bit (but not too much!) for a new 512 DIMM, and when I opened the box and smelled the chip smell, put it in always worried I was going to fuck it up, and then computer literally felt faster that next boot...that was pretty fun!! Now it's like oh great $5k for a slab of stone that can do pretty much anything, neat. I still think computers are cool, just not particularly exciting.

Me either. I guess it's just fatigue, at least for me. I also don't really get that excited by new LLM releases either. Not to say the tech isn't impressive, but I guess all the hype has me inured.

Because it's the same shit every year for the past 5 years with the M line. 2010 to 2015 was a major improvement, 2015 to 2020 was a major improvement, now they pretty much solved the computer/laptop problem for 99% of people. I'm on a 16gb m1 air, I see absolutely no reason to update.

Because the M1 was too good, a qualitative leap over previous Macs and really every other laptop and even some desktops back in 2020. Now, Apple Silicon is just iterative.

Not true everywhere. Only where required by law, so complain to your government.

Why would you want to do that? Do you like the hardware that much, and also that much more than just an M2 (soon M3) running Asahi?

Linux in a VM would work with the usual caveats. Periphery like the built-in webcam most likely won't work. Getting codecs and DRM to run will be pain and you'll be back to use macOS for that quickly (but that's just standard pain of ARM Linux).

It's not $200 across the board. M4 MacBook Pro and M5 MacBook Pro started at $1599 with 512GB SSD.

Now it starts at $1699, a $100 bump but comes with a 1TB SSD. Previously it would have cost $1799 for the 1T SSD, so it's a $100 bump on base price but you are also getting 1TB SDD for $100 less than before.

I was considering it but got cold feet when I've been told that you could damage it when cleaning it. When I open/close my laptop I leave a ton of finger prints. I'm not too good with delicate hardware stuff.

With an additional $200/month subscription from Anthropic, because they noticed that the Kimi K2.5 they were able to run on their M5 comes nowhere close to matching Opus 4.6.

Same, in fact the only reason right now that I would upgrade my m1 pro is if they threaten to change the design by getting rid of the hdmi or sd card slot, or doing something stupid like when they added the touch bar. I was locked into my old intel pro for so long because of all the bad hardware choices they were making.

I read it the same way. I should've gotten way more RAM back when I got my M1 and RAM was still cheap although this was of course before the LLM boom so there was no way to really know.

Well, I just upgraded from Intel late last year. There are lots of users still on Intel :)

My 32gb m1 max was probably the best purchase I've made. Still plenty of headroom in performance left in this beast. Wonder what reason they'll use to end software support in the future. Bet it'll be some security hardware they make up for the sake of forcing upgrades.

> NVIDIA RTX 5090 already offers 1,792 GB/s

You can buy two m5 pro base model for the same price as a single 5090...

You're right a $3600 graphics card is worse than a $2600 laptop; but from my perspectives they're very different products. Not least of all because even at $3600 for a RTX 5090 you still have the whole rest of the computer left to purchase.

I imagine the upcoming M5 Ultra will be competitive in this regard. The M3 Ultra already has 819GB/s and it's two generations behind.

M5 has been out since last year, no?

Not sure either since M5 base has been available for months now

This is always the gamble with buying a Mac. Either purchase right when the new is released, or be on the fence of your new becoming old a couple of weeks after purchase.

Why do you think the price went up by $200?

That's a big part of what's keeping me from upgrading. Every time I look at my wife's iPhone I'm dumbfounded by just how bad the liquid glass looks.

It's the first time I've ever been so repulsed by a design that I actively avoid it just... out of sheer preference.

I have a base M5 since last year. You cannot, no. It is literally impossible. Do with that what you will.

It does look terrible, but I haven't found it to be unstable, personally

accessibility settings can turn off some (but not all) of the garish animations, transparencies, etc.

You barely see any liquid glass on Tahoe. I keep my dock hidden and it's just the icons mostly which aren't that different than before.

Good point about the telemetry part. I've been using Little Snitch for the past few years and just block all the telemetry calls.

Or don’t use Google, Adobe or Microsoft software if that bothers you? And how is that Apple’s fault?

It would probably be worth finding a more friendly way to market this, but it's a reasonable / accurate way to say it.

The prompt processing sped up.

Not the output generation.

M4 was notoriously slow at this compared to DGX etc.

Or just don't but an Apple charger? You can get a perfectly fine small 100W GAN USB-C charger for like $30 on Amazon.

Since that is required by law, I suggest moving.

Tell me about it. I checked the page thinking whether I should go for 256 GB or 512 GB RAM model.

128 GB maximum.

Sigh.

I'm assuming you're in the EU or UK, Apple is required by law to not include a power adapter:

https://appleinsider.com/articles/25/10/15/eu-gets-what-it-a...

In the US they provide one in the box free of charge.

This is one thing I don't really blame Apple for, and I think everyone else will follow suit -- and not just because Apple is doing it.

The EU requires that users must be able to buy a device without a charger. It's a huge supply chain challenge to add two variants of every single SKU, one with a charger and one without. So the obvious solution is to sell the charger separately, since you need that regardless, and always sell the device without a charger. You avoid having two variants of everything that way.

Now, you could maybe argue that Apple should default to bundle a charger with your laptop, so that you'd have to uncheck a "bundle charger" checkbox on their website. But do you really care whether your laptop costs $2200 and you can buy a charger for $60 or your laptop costs $2260 and you can save $60 by removing the charger?

You can make an argument that doing it Apple's way hides a price increase. And yeah, that's probably fair. But it's not like Apple is afraid of non-hidden price increases either.

I have a huge tote box full of power bricks, most of them white Apple ones. I have a stack of 60-90W Apple USB-C ones too that I don't use cause they only have one port and are larger and worse than modern GaN units that can do 140W on one port while also pushing 30 or 60 on the others.

So, if you want one of mine, you can have one. On me. Because I'm fucking drowning in the things and appreciate not having to deal with another one.