I was blown away by M1 Pro, and used it for 4 years before it became a little sluggish, and replacing it with an Asus S16 running on Ryzen AI 9 HX 370 (an excellent laptop). One of the main reasons was missing my old Linux setup.

I've briefly tested M5 Pro in an Apple store and was surprised by how quick it felt and did anything. A tangible and significant difference, and I really feel it would be good getting it, or M6.

However - before, MacOS was a big factor driving me towards Apple - now, it would be hard for me to give up my Linux setup and all the stuff I love about it, even for such performance...Linux progressed really nicely, while MacOS deteriorated at the same rate, and it's changing the balance and the decision for me.

I know, this is a bit of a meaningless comment, but it's funny in a way. Feels like late 90s again:

    - Xiaomi: We have matched Apple in CPU performance.
    - Apple: *Meep Meep...*

Even w/the pricing spike, inflation adjusted we are back to roughly the prices of a new Mac SE/30 for something that can beat a turing test w/o sweating.

I yield the floor to no one when it comes to pessimism, but that's incredible.

Rumors say that Apple will only release M6 base variant and skips M6 Pro, M6 Max and M6 Ultra variants to concentrate all efforts to create a good AI capable M7:

"According to reports from Bloomberg, Apple will be skipping its M6 Pro, M6 Max, and M6 Ultra chips to accelerate development of the M7 chip. That means the only chip to be released from the M6 family will be the base M6.

The reason for this break with tradition: AI. Apple had been planning major neural-processing upgrades for the M7 family and ultimately decided those improvements were important enough to justify accelerating the next generation rather than completing the M6 lineup." https://9to5mac.com/2026/08/08/apple-m7-chip-heres-why-it-ma...

I'd skip M5 and M6 chips for LLM work and wait for a year for M7.

Apple Studio with maxed out M5 Ultra, 256GB RAM and 16TB storage is 18,299$. The 512GB RAM version apparently is coming in October, considering that the difference between 96GB and 256GB is priced at 4000$, the 512GB upgrade must be eye watering.

So, on the mini the RAM upgrade runs at 25$ per GB on all tiers, the same as the Studio therefore the upgrade to 512 will probably cost 6400$.

The fully maxed out Apple Studio then will be 24699$. It's 17199$ if you don't upgrade the storage(1TB).

Nevertheless I itch to have one :)

Apple still has the best hardware so I moved to it for the last few years, but the closed software ecosystem is terrible for taking advantage of it.

I wasn't able to debug network errors (restartin my Mac worked), Metal was missing low level disassembly / debugging tools (there is some hard to use UI), but the worst thing was the inflexible windowing system.

Even getting all the window handles on all screens/desktops with their titles and programs is impossible.

I just decided that I move to Omarchy 4 (basically Hyperland + QuickShell) + NVIDIA GPU, and I already was able to customize it more than my Mac in years.

I will miss Apple's hardware for sure, but not MacOS and the missing hardware documentation

More validated by the day that my $450 M4 Mac Mini (16GB) was the best deal in computing for a long, long time.

Add ARM boot camp, MacOS is too restrictive compared to modern Linux. Asahi's devs can't keep up with the constant hardware releases so official support is important.

The 512GB Ultra is amazing, sure. But who is it for exactly? VC funded big spender founders? In that case why would they need local AI? The ultra rich enthusiast? But there can't be too many of those. So who actually buys these?

>fluid frame rates in demanding games like Mixtape.

not getting on that bandwagon but wasn't that not the most demanding game as its a just a nonstop cutscene.

Leasing now an option, only $50/month (cheaper than inference subscription?), so even cash-poor can go the amortized-investment route.

I've often felt there is tremendous value locked up in underutilized old computers. It would be interesting to see Apple in 3 years offering compute as a service using lease returns (or more likely, partnering with someone else to operate it (perhaps exclusively in secondary markets like China or India, to address political demands for local siting or jobs). Apple is in the best position to work around or even gap-fix older software/hardware limitations in a controlled environment, and now they can do so without cannibalizing new hardware sales.

I run a lot of local models (I am always experimenting) on my 32G M2-Pro MacMini - I would love to upgrade.

The financial aspects don’t work however: I can learn and experiment with what I have for local models, and I pay as I go on FireWorks.ai for open model inferencing and no matter how much I use this service my monthly bill is between $10 and $40 and much faster than any reasonable home rig.

Hybrid ‘small local’ and buying inference is the way I choose.

I somehow find it better to give 2 frontier model companies 100-200/month than dropping 10 grand on a hardware that will get old in no time with bad TPS. I really want to have a fully local model but seems like one more generation wait and we will be there?

I wonder to what extent the timing of this release was driven by their need to increase prices, but not wanting to apply it to in-market hardware. It's kind of a weird time to drop new Macs, between WWDC and the iPhone release, but not early enough that it could be purchased by college students. I get that most students use laptops, but if I were heading off this year I might use an iPad + Mac mini for more horsepower while maintaining portability.

I bought a 128GB M4 Max Mac Studio a while back, and for a while I thought like I had done really well to buy it when I did.

The problem I'm having now is that no models are targeting RAM of that size. Everything is either much smaller, targeting laptops, or much larger, targeting hardware well out of reach of enthusiasts.

Please, AI people, start making models targeting 128GB machines again. The last interesting one was Qwen 3.5 122B.

Why do they focus so much on hardware when their biggest problem is software? I don't get it. My Dell XPS 13 is probably not that powerful in terms of performance but Linux is so much more productive. This drives me crazy

I feel like the M series macs have been some of the most positively received computers ever made. I am yet to meet someone who doesn't love theirs, myself included.

96GB -> 256GB upgrade costs 4000 GBP in UK or $5460. $34 for GB.

In US its $4000 upgade so $25 for 1GB.

Also:

> 512GB memory option for M5 Ultra coming late October

Just bought AMD AI 395+ 64GB RAM for 2000eur. Mac mini would be 2x for about similar performance. Pass.

I was just comparing this to an rtx6000 96gb build and when the f*ck did nvidia double the price?

If you want to comfortably afford this gen you had to trade options on memory stocks...

The page says 170G/s memory bandwidth for the NPU and 1.2T/s for the GPU. Why the discrepancy if it's all "unified memory"? The former is nothing to write home about as far as AI compute is. The latter is really nice.

Which one is it you can run local models on? I suppose the NPU only.

I have a m1 pro macbook, its still runs very smoothly. Moving from Intel to M series chipsets were a game changer.

I like my Mac mini as much as the next person, but I don’t think an M5 ultra 64gb with 4tb should be as much as a DGX Spark with 128gb and 4tb.

Can anyone recommend the perfect sweet spot for someone who wants to run their own inference?

I would love to see real LLM performance benchmarks for these machines. Apple statement regarding LLM performance seem little vague.

I still struggle with local llm being a reason to drop insane money on one of these - they are targeting people who want high performance local LLMs, that value privacy over using 3rd party api, but also dont want to use Nvidia? small part of the pie by any reckoning?

This feels like paying a lot for something that is just not very good yet - cloud models are too good vs local models (may change in the future..)

I think I remember that at the 8nm architecture times there were talks about "reaching the physically possible limit" with the size of the transistors.

These are 2nm. While normally pretty cynical, I am experiencing one of those "what a time to be alive" moments.

I would love to buy the machine but I hate Tahoe.

FWIW scaling up from https://huggingface.co/avlp12/Qwen3.8-27B-Alis-MLX-6bit and some other sources:

You might expect the M5 Ultra to produce 50 t/s from Qwen 3.8 27B with a good context length.

My macbook pro M1 is still running fast and silent.

Damn. I just bought a maxed out MacBook Pro M5 Max 128GB 8TB, still waiting for it to be delivered. I could get 256GB RAM M5 Ultra 1TB for roughly the same price, and it's double the memory bandwidth. Which one would you recommend? I do plan to run local LLMs.

Sick. Particularly stoked for the 10gb network card ($100 option) when using the Mac Mini as a server. Just wish the memory + NVMe prices could come back down to pre ai-goldrush prices. As $2999 for the M5 Pro with 64GB RAM feels painfully over-priced.

Does anyone know if they’re actually at 2nm transistors, or if this is more marketing talk? It’s kinda like when TVs advertised their contrast ratios (do they still do that?) but they all used their own measurement methods, so dynamic contrast ratio became a worthless metric.

Do you think that little moniker will just disappear in a decade or so as we reach 0nm advertisements, or do you think they'll just add more decimal places? 0.0001 nm! Next year, 0.00001!!

Nice! Linux Please.

My M5 MacBook Pro, while amazing, is basically relegated to being a thin client with amazing battery life that exclusively SSH's into my desktop PC for work.

I could spawn a Linux VM / use Docker/Podman locally, but dealing with allocating memory, hard drive space and whatever other limitations non-native containerisation incurs isn't worth the squeeze.

Plus I really want to have Vulkan access, both so I can actually play games like a sane person and also use standardized local LLM tooling (MLX is cool but I don't like dedicating time learning things that I can't leverage elsewhere).

If the Mac Mini / Studio range had Linux support, I might actually consider buying one for my home lab. I know others already do and just put up with MacOS - but if I buy a "pro" product, I want it for "professional" usage damn it

Great devices for watching YouTube and video editing I guess?

Tangential, but what would be the ideal Mac option for home movie editing, casual gaming and amateur CAD fiddling in Fusion?

I plan on maximizing my residual student benefits, and taking advantage of education pricing.

I did a little stint studying thermal cyclic fatigue in multi-die packages (many years ago) and I'm wondering how the issues there are solved these days.

Was the quad die layout expected? I've checked a couple articles and can't find any commentary on it, but I don't remember hearing this rumored.

I feel so left behind with Linux machine on x86.

It's rather interesting that Apple (and Arm) are blazing the trail while Intel is seemingly left in the dust. Maybe not entirely on raw performance, but definitely on performance per watt.

Usable ram amounts in late October

How many Neural Accelerators does M5 Ultra has in its 64 vs 80-GPU variants?

Anybody knows?

Just to clarify, Neural Accelerators have nothing to do with Neural Engine.

Interesting that "coding" is now part of the marketing brochure as one of the use cases, while that was historically kind of missing. Is this new?

0% APR for 12 months (24 for iPhones only?) from Apple Financial Services for a device that can approach or even exceed what we were paying for new cars just a few years ago. Apple is definitely making bank off these financing offers, and with very little risk as unlike a car these Mac Studios don’t lose 20% of their value when you drive them off the lot.

Interesting times, to say the least!

So many people talk about wanting Linux on Apple Silicon, and while Asahi can run on M1/M2 and has some support on M3, it's not optimal. Also, software quality in Linux, while higher than most alternatives, has been trending downward. I am more eager to have NetBSD's Apple Silicon support improve than I am Linux's.

https://wiki.netbsd.org/ports/evbarm/apple/

do 4x512gb builds actually make sense now to run a big model (beyond a cool demo)? or are 200-500B moe models the sweet spot? In which case clustering 256gb builds might make more sense.

~12k for 80 core gpu with 256gb, 14k in October for 512gb. Seems like that could make for a very descent on prem inference server.

Gonna go sell a kidney, should just about cover a base Mac Studio. Guess I'll need a payday loan for the power cable

Curious how the M6 neural engine compares for local LLM inference. That's the real benchmark now.

So the idea of networking multiple Mac Minis via RDMA over Thunderbolt 5 to run local LLMs is intriguing, but doesn't really seem reasonable at this price point? What are the more cost-effective alternatives?

The M6 is useless for AI. Is there any model which is actually useful and fast on 32GB?

All I wanna know is how many tokens per second on 100B MoE??

Have we cracked 100 yet.

When we crack 100 tok/s at 2M context for <$10K I’ll buy one.

I'm still using M1 pro, which still runs buttery smooth. No complains but yes 16 gb ram feels little less in some tasks.

What's everyones recommendation for one to run a good local LLM model on?

Does anybody know how much AI was used to build this? The release cadence that Apple has with its CPUs is impressive.

If I compare to like, January 2024, the prices for RAM these days make me want to weep.

"M6 features a powerful, larger 12-core GPU that provides higher geometry rates and updated Dynamic Caching to deliver stunning visuals and fluid frame rates in demanding games like Mixtape."

Why choose the worst game of the year? Just because it's made by the daughter of Larry Ellison?

I used to think Apple was ngmi but turns out they didn't have to do anything new and still win

Maxed out Studio is $30,000+ tax in Canada if financed through Apple.

That's wild!

M5 Pro in a Mac Mini with 64GB RAM and 10Gbit Ethernet seems like the perfect Jellyfin server and Ollama test server. All for just over $3K (I specced with only 1TB local nvme).

Wouldn’t it be amazing for Apple to give us a MacBook Air 15” M6 with a 15W sustained passive TDP capability?

Just amazing engineering push, the competition got the message and we benefit.

I have Mac M1 Max and I'm quite happy with it. But these advances make me think that maybe I should upgrade to Mac M6 (or something) when it is released.

My M2 still does everything I need, but the M6 is seriously tempting.

Will forever miss the 499$ mac mini!

Hmm, I was about to get a MBP with an older Max chip, guess I should wait for prices to slide a bit.

This fits neatly in the backpack with the MacBook Pro, if I need more compute power.

Is this the best bet for running local AI now? The choices are overwhelming.

A Macbook Neo with an M6 and 16 GB RAM at $699/€699 would be a killer feat

cool but maybe i was expecting much more from both m5 ultra and the next gen so it feels somewhat underwhelming. Hopefully m6 max is more than just a 10% improvement over m5 max

A little expensive, but very powerful.

When will Apple's Mx CPUs use Intel's 18A/18A-P/14A node(s)?

My m2 max is still running strong, but may upgrade soon

I had a look at RAM prices lately. Two 32GB RAM I bought about 3 years ago, now cost exactly 300% as they did back then. AI companies owe us a lot of money here. They drove up the prices. They should pay us the difference.

Very well timed for John Ternus's first quarter.

Saw a 768GB RAM mac coming soon, would wait for that.

I'm so disappointed with the memory pricing we have to endure... MacStudio M5 Pro is $2,499 base with 36GB RAM, but when you increase it to 256GB, the price jumps to $9,499!!!

...Additionally, M5 Ultra features a massive amount of high-bandwidth unified memory, up to 512GB, and delivers a staggering 1.2TB/s of unified memory bandwidth that is 50 percent higher than M3 Ultra.

That's faster than majority of GPUs available today: 5090 ~1.8TB 5080 1TB 5070 0.72TB My R9700 0.64TB :(

A computer created around a GPU and/or a shared ram to run LLMs should not be called a computer anymore. It should be called an inference engine. I'm not against it but it could be stripped down to reduce some cost and it would be the perfect companion in a lab.

If only their stuff could reliably run Linux on the release day.

With 512GB of RAM, it may finally be feasible to run both Slack and Teams.

How much did Larry Ellison pay to have his daughter's pet video game project featured on there?

When is the m6 air coming though

i wish i could use their hardware, but i refuse their software

Every time Intel/AMD gets close enough, Apple just crushes competition in terms of SoC. I don’t know who’s on that chip team but they’re world-class. Hope they get paid more than a bunch of AI idiots Meta hired to do absolutely nothing (but I know they are not even close).

If Apple does design a RISC-V frontend to their MX microarchitecture... ARM better keep those ISA license fees close to zero.

look at the prices !! crazy

Apple is a terrible company through and through that should have been broken up million years ago. But, nobody dares to question apple.

My 2019 macbook pro recently started giving me CATERR errors. There are a few forums and LLMs tell me it is logic board. I took it to Best Buy - a location suggested by apple support and they say we could replace battery but there is no guaranteee it will work. Logic board replacement would be $600 to $800. Unfortunately, only Apple can repair it given how every component is integrated and soldered. There is no help available from Apple support or anyone on this issue.

Apple creates "Planned Obsolescence" deliberately. Every year they launch new products with minor improvements and that basically distorts markets. There is MEGA scalping going on for Mac Mini but Apple does not care. As a trillion dollar corporation seeking profits Apple does nothing to help price gauging. Unfortunately, there is no competition. Microsoft is 10x deep into its own issues.

I wish we had startups who came up with strong competition for Apple.

This is weird. These options are very different from existing options. The M6 memory bandwidth only went up a little bit, which means the bigger AI compute is not going to be worth much in practice. A base M6 should only be close to matching an M1 pro (and it's not even close to an M1 max)

M1 pro: 16 GPU cores, 200GB/s memory bandwidth

M6: 12 GPU cores, 170GB/s memory bandwidth

Now the AI compute is over 5x more. So it should probably be possible to make models that perform better on the M6, but the same models just won't work.

But the M5 pro mac mini will wipe the floor with the base M6, as most of its stats are double that of the M6 chip. Should be double the performance easily.

Perhaps also relevant: the M5 pro mac mini should have very similar to double the performance to the Nvidia DGX Spark, assuming the model fits the memory.

So that is probably the chip it should be compared to for a quick guess about benchmark results.

"in demanding games like Mixtape"

I was going to stop making low effort comments, but, come on, you have to agree Apple started making those first.

> M6 supports up to 32GB of unified memory to multitask across demanding apps

I can't believe that Apple still comes with this bullshit like 32 GBs is a lot. It's a lot for video memory - vRAM, but not RAM.

So I had just bought from Amazon in July an M4 mini, brand new, for a lucky post-hike price of 832 CA$ (682 USD); it was the very last unit at that price and I snatched it the morning it listed on the Apple/Amazon.ca official vendor. So far I've been using it for a month and it's been great (switching back from a Windows desktop for a decade).

Today I entered this new M4 mini's serial as a trade in for the M6 base mini, and the retail price went from $1249 CAD to $824 "once trade-in received". That's the automatic estimate, so basically I'd be paying twice for an upgrade...

Super psyched for the upcoming Mac Mini. Maybe the old NUC can be tossed.

Call me when Linux will be officially supported. After that I can look and see what else can I get for the price and maybe then ...

yesterday someone posted a link saying xiaomi "matched" apple's latest M series performance. Was that for less than 24 hours?

Is Apple just going to announce everything silently from now on? No more getting excited for the big events to see what's new -- it just appears on the blog one day?

Also: "a staggering 1.2TB/s of unified memory bandwidth" -- yay, the GPU has reached the year 2020! (I'm a bit bitter that my M4 Max is near useless for local LLMs because of its low memory bandwidth.)

I should have bought a 100 m4 mac minis when I had the chance. Thanks hyperscalers for buying all the supply and renting it back.

A m4 mac mini is better than al of these per dollar, msrp adjusted.

Hopefully by the end of the decade China figures out manufacturing at scale and fixes this.

Why do I only have two kidneys dude

will be great fun if one M5 Ultra with 512GB memory at 1.2T bandwidth capable of doing 3x smallish local model inferencing each at Opus 4.5 level of intelligence.

Well said

> M5 Ultra features a massive amount of high-bandwidth unified memory, up to 512GB, and delivers a staggering 1.2TB/s of unified memory bandwidth that is 50 percent higher than M3 Ultra.

Apple never needed to participate in the AI race to zero. Because they were already at the finish line years ago building their own chips that can run large >100B parameter AI models locally.

"hacker news"

I wonder why they even bothered with m5 ultra instead of going m6 ultra directly, honestly.

What’s the point?

Now if only I can natively run Linux on their hardware.

> Apple’s developer frameworks and tools — including Core AI, Core ML, Metal, and Xcode

Can we please kill the xcode. It is worst pile of garbage I have to use just to develop ios app.

fucking awesome

[flagged]

[dead]

[flagged]

[flagged]

[flagged]

[dead]

[dead]

[dead]

[dead]

[dead]

>up to 1.2x faster multithreaded performance as compared to M5

Massive. I feared it would only be up to 1x the speed of an M5 /s

[flagged]

>M6 supports up to 32GB of unified memory

Is this a joke?

>Additionally, M5 Ultra features a massive amount of high-bandwidth unified memory, up to 512GB

Now we're talking. But at what cost?

512GB late october sounds lame.

and 512GB is so 2025.

Give us 1TB version. Where is the competitive spirit?

Cannot believe "moar transistors" is still the only idea they have.

From what I’ve noticed, Apple products have been getting worse in quality year after year. Sometimes they even ruin their own devices with updates... I guess it’s all because of marketing.

I know they are a phone company, but I think they should focus on local models software, not only hardware.