We're getting close to local AI being really useful. Before the M1 is dead we might have already seen a huge shift. In my mind, there are two types of macs: those that have the bandwidth to run local models (Max & Ultra) and those that don't (Base & Pro chips).
The generation doesn't mater as much to me as the memory bandwidth. When that day comes, I'd rather have a M1 Max 64GB over a M5 Pro 64GB.
Local AI already is really useful. This has been true since Qwen Flash Next and Qwen 3.8 27B rolled out. The struggle with exactly 64GB of unified RAM is that you end up having to stream the engram table from SSD because you will need almost all of your 64GB to run QFN. A 27B dense model still runs a little slow, even on an M1 Max Mac. However if you have even an older gaming GPU with 12-16GB of RAM and enough system RAM you can run this model (QFN) at a very respectable token rate and intelligence level. It is a Sonnet 5 class model you can run at home. There are new quantization techniques that make these models very good at agentic tasks.
And now if you stretch you can run DeepSeek Flash 4.1.
And if you were willing to roll up your sleeves you could build your own AI lab at home and run this model and models like GLM Flash 5.3 and DeepSeek Flash V4.1 at lower quants. The budget GPUs (like V100) are kind of not budget now, but still <5K to acquire 128GB of fast VRAM that can compete with a 5090 and actually has a lot more interesting options available IMO.
Can share a good way of running Qwen 3.8 27B (or similar) on 16GB VRAM? I've been trying but I end up with either a heavily quantized model or a small context or a slow speed.
Dense models are not well suited to this. Depends on your system, but most inference engines let you put the KV cache on the host. They should let you use "Q8" KV cache format as well which takes half the storage of bf16, the default format. You essentially are stuck at having to fit the entire dense model into VRAM because it must compute across the entire model every token. Without it being in VRAM it is slow. MoE models such as Qwen 35B with 3B active parameters are much friendlier to limited VRAM systems. This is why you see systems with 16GB of VRAM running something like DeepSeek Flash if they have a bit of spare system RAM. Usually not at any good token rate, but tolerable.
Dense models also don't like quantization as much and are less tolerant of it. There are some very interesting smaller models like https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4... Dory that I would check out. You need system RAM or to stream a larger MoE model from disk and it won't be fast. You need to specialize the models some to get quality results if you have very limited system resources.
My setup is a Ryzen 3600, 90GB DDR4 allocated to a VM (don’t even really need that much, just had it available, special thanks to the year 2024), and an RX 9060XT 16GB ($500 GPU).
I get around 40-60 tok/s depending on the quantization level I choose.
As you can see, the CPU is nothing special, don’t even need DDR5.
The easiest thing to do (takes 5 minutes) is to use LMStudio and then search for the model inside LMStudios UI.
It will show you on the model selection IF a model fits and if you use a "Qwen 3.8-35B-A3B" type of model, it should just work out of the box. A3B stands for Active which means you need a lot less fast ram.
You will run into issues fast though: If you have big tasks for it to do, your pc/mac will get warm very fast and noisy and you will cosume relevant more energy.
AI harnesses like Claude will set it all up for you, you can just say "I want to run a localllm on this machine" and it will find the best one for your hardware and even help you fine tune it etc
You will end up with a sub-par result. The space for small models is more specialized and specific and you should optimize to the specific task and or tasks you want to do. It will get some local model working, but there are a lot of interesting small models in the 4-10B space to look at and play with for your specific tasks. If you expect a full agentic coding agent it's just not going to happen without a larger model like Qwen 35B and ... that is not that great by today's standards, even Muse Glimmer is better.
That's only part of the issue - prefill is horrible prior to M5 class chips.
So it depends on what you're doing. Using it for a fun oracle without feeding it too much context per turn? Sure, an older Max chip is great.
As a coding agent doing meaningful work on a sizable project? Compared to a cheap as chips cloud option, it will take an order of magnitude more time to complete.
Puh i don't know, there might be plenty of new/better chips coming out for AI next year or in 2 years.
Isn't it also rumored that the M7 is reworked for AI?
And Xiaomi announced their AI Cube with 1.2t of bandwidth. Just because Nvidia is protecting its business by giving consumers either compute or bandwidth and Apple being something in between, doesn't mean we can't get both a lot better balanced.
Yeah, the price is so high and things are moving so fast I don’t wanna risk committing to something that becomes obsolete in a year. I wanna see the rumoured M7 first. Hopefully hardware demand doesn’t make waiting a stupid decision…
>The generation doesn't mater as much to me as the memory bandwidth
I don't know man?
The generation kind of does matter. I'm not gonna lie, matrix multiplication units speed things up considerably. That hardware confers a material change in capability.
Maybe from M5 up, it doesn't matter? But you're definitely going to notice the difference between M1 and M5 for instance.
Still on my M1 Max 64GB from release date. Love it. I don't do local AI but this laptop runs so cool and fast I have no reason to upgrade. They really knocked it out of the park on the design.
It feels just as fast as day 1 and I've never had any speed issues. Say what you want about Apple and their pricing and ecosystem but I've been using their hardware for over 20 years now and it's always been top notch.
If you buy their support/AppleCare, your support and warranty is excellent too. They don't skimp. I've seen them go so far past the extra mile it's unreal.
This is also my experience, but when I saw how few tokens/second with a large model it gets locally I was bummed. I’m wondering if I’ll be obligated to upgrade in the near future
> I'd rather have a M1 Max 64GB over a M5 Pro 64GB.
I am one of those: bought the M1 Max with 64gb thinking in running VMs, afterward bought a lighter Air with less memory and ended up using the M1 Max as a kind of luxury for LLMs and multicore computing.
Considering the insane memory price surge it's going to take years (or an eternity) before most people will consider upspecing their machines to run local over paying $20 per month to have the honor to loan a slice of Altman's hardware.
The only hope we have is the AI bubble bursting, but that will undoubtedly come with its own set of macroeconomic surprises.
The ~$1T/y AI capex isn't remotely sustainable. Memory fab capacity will go up, AI datacenter provisioning will go down, prices will become reasonable.
Though I have had good results buying 1-2 generation old refurbished Macs as my secondary machines or for family members who don’t need cutting edge everything.
It’s even more valuable now that Apple silicon of every generation is more than enough to run every non-power user use case short of newer video games.
They seem to want those who would’ve bought a high spec iMac (or briefly, an iMac Pro) to instead buy a midrange-or-above Mini or a Studio instead. With the M-series, the iMac returned to its roots as the “family computer” or “grandma email box” kind of machine.
It’s too bad. A high end iMac still holds a lot of appeal, especially for those seeking clean desk setups. I’m using an MBP hooked up to dual monitors through a dock right now and it’s fine, but I do miss how clean my desk was back when it was a VESA mounted iMac Pro and companion monitor.
I was just thinking the same thing. The M-series is so good, the demand for uber-desktops is smaller than ever. And all professionals I know truly want/need the portability of a laptop.
the iMac returned to its roots as the “family computer” or “grandma email box” kind of machine.
My preferred setup would be a semi-beefy desktop paired with a highly mobile laptop, which with the current product line would be something like a Studio (or a high end 27”/32” iMac, if such a thing existed) and 13” Air. I ended up with an MBP due to constricted availability of the Studio/Mini, though.
I feel like in 2026 maybe they realized that "pro" users (engineers, designers) mostly just want laptops + external monitors anyway. The Apple Silicon CPUs are so f'ing good, the % of people who need an uber-desktop is smaller than ever.
24" is a good size for casual users which is the main market. You can have two roomy-ish GUI apps (two web pages, etc) side by side.
The M6 Pro and Max variants with higher memory support have been dropped in favor of accelerating M7.
This has been reported for many months. Sorry if you missed the news, but if you need a much higher memory variant the M5 Pro/Max are what’s available this cycle.
The panel is incredibly thin, but the base is a few inches, and the amount of space it will take up due to tilting the display is also a couple of inches. After a certain point, thinness is a meaningless aesthetic choice that serves no function whatsoever.
Thinness has been frivolous aesthetics since Steve Jobs times. A thicker laptop will always mean better thermals, more silence, more battery, repairability, more ports.
I wonder why the iMac still has that chin design and small panel. They’ve been producing displays for long enough that the iMac and Studio Display should have the same form factor.
The chin is where Apple puts the logic board and speakers for the iMac. Yes, they could move that behind the display, but I'm guessing that'd make the display thicker (which they wouldn't want) and make the thermals harder. It's certainly something they could engineer, after all the iPad has its logic board behind the display. But it's not like the chin serves no purpose and it probably helps with the thermals and sustained performance and probably makes it easier to engineer everything for what is probably one of their lower volume sellers (compared to iPhones, iPads, MacBooks).
I worked for an AASP in the early 00s and I questioned this, considering that previously the logic boards were actually housed behind the display for both PowerPC and Intel models[0]. This does appear to be mostly true[1] (some of the logic board pokes up beyond the chin).
It'd be interesting if they used that power brick tether to drive the display similar to the Samsung Frame, which takes power and signal from a fiber optic cable that runs really easily in a wall. If they could shuttle Ethernet up the power cord, they could send video signal back, too.
I love the magic mouse, had one going since 2016 and it is still great. Not a fan of their keyboards but the swipe + scrolling exp. on the magic mouse is great IMO.
The maximum of 32 GB RAM has to be a joke. My Linux laptop has more RAM, more SSD, more CPU at the same price, and it can run games on top of that. Hard pass, even if I were to tolerate the OS.
I read similar reports so many times, when I got my 165hz screen I got disappointed. I also disable it on my phone to get more battery life. Different folks, different strokes.
I switch between a variety of displays throughout my day, ranging from 60hz up through 240hz and yes there’s an obvious visual difference, but I adapt to it in a matter of seconds.
It’s nice for games but I don’t really care for productivity use. In fact in that situation I’d take integer-scaled HiDPI density over refresh rate in that scenario.
How many watts are you saving with that, realistically? If that few minutes of extra battery is a game changer, you need a bigger battery, not a worse screen.
> Just like Intel not shipping ECC for client SKUs when DDR5 design makes it really necessary
The ECC in DDR5 is on-die ECC and that's not "real" ECC, it's what DDR5 needs to not fail. Side-band ECC is the standard method, employing extra memory chips and 72bit memory. OEMs mix these in the marketing materials to fool the users into thinking they buy the better option.
Yes, here I'm referring to actual ECC, not the on-die ECC. All DDR5 chips have on-die ECC as a prerequisite just to reach error levels comparable to standard non-ECC DDR4, but from a hardware analysis perspective, even with the on-die ECC, DDR5 still needs ECC more than DDR4 (which already needed it).
I still think those expensive Apple monitors (which even had RAM) were intended to be 'iMac Pros,' but due to stagnant growth in the all-in-one market, they never saw the light of day.
The generation doesn't mater as much to me as the memory bandwidth. When that day comes, I'd rather have a M1 Max 64GB over a M5 Pro 64GB.
And now if you stretch you can run DeepSeek Flash 4.1.
And if you were willing to roll up your sleeves you could build your own AI lab at home and run this model and models like GLM Flash 5.3 and DeepSeek Flash V4.1 at lower quants. The budget GPUs (like V100) are kind of not budget now, but still <5K to acquire 128GB of fast VRAM that can compete with a 5090 and actually has a lot more interesting options available IMO.
Dense models also don't like quantization as much and are less tolerant of it. There are some very interesting smaller models like https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4... Dory that I would check out. You need system RAM or to stream a larger MoE model from disk and it won't be fast. You need to specialize the models some to get quality results if you have very limited system resources.
https://github.com/Niko1221/Strata/
My setup is a Ryzen 3600, 90GB DDR4 allocated to a VM (don’t even really need that much, just had it available, special thanks to the year 2024), and an RX 9060XT 16GB ($500 GPU).
I get around 40-60 tok/s depending on the quantization level I choose.
As you can see, the CPU is nothing special, don’t even need DDR5.
It will show you on the model selection IF a model fits and if you use a "Qwen 3.8-35B-A3B" type of model, it should just work out of the box. A3B stands for Active which means you need a lot less fast ram.
You will run into issues fast though: If you have big tasks for it to do, your pc/mac will get warm very fast and noisy and you will cosume relevant more energy.
And as you mention: yeah small context!
Its still not that practical in my opinion.
So it depends on what you're doing. Using it for a fun oracle without feeding it too much context per turn? Sure, an older Max chip is great.
As a coding agent doing meaningful work on a sizable project? Compared to a cheap as chips cloud option, it will take an order of magnitude more time to complete.
Isn't it also rumored that the M7 is reworked for AI?
And Xiaomi announced their AI Cube with 1.2t of bandwidth. Just because Nvidia is protecting its business by giving consumers either compute or bandwidth and Apple being something in between, doesn't mean we can't get both a lot better balanced.
I'm gonna max out most of the specs on the next one I get, would feel insanely wasteful to do that now.
I don't know man?
The generation kind of does matter. I'm not gonna lie, matrix multiplication units speed things up considerably. That hardware confers a material change in capability.
Maybe from M5 up, it doesn't matter? But you're definitely going to notice the difference between M1 and M5 for instance.
It feels just as fast as day 1 and I've never had any speed issues. Say what you want about Apple and their pricing and ecosystem but I've been using their hardware for over 20 years now and it's always been top notch.
If you buy their support/AppleCare, your support and warranty is excellent too. They don't skimp. I've seen them go so far past the extra mile it's unreal.
Definitely grabbing 128 GB + for the next machine.
I am one of those: bought the M1 Max with 64gb thinking in running VMs, afterward bought a lighter Air with less memory and ended up using the M1 Max as a kind of luxury for LLMs and multicore computing.
The only hope we have is the AI bubble bursting, but that will undoubtedly come with its own set of macroeconomic surprises.
Though I have had good results buying 1-2 generation old refurbished Macs as my secondary machines or for family members who don’t need cutting edge everything.
It’s even more valuable now that Apple silicon of every generation is more than enough to run every non-power user use case short of newer video games.
I would love to have a pink one for my MBP!
It’s too bad. A high end iMac still holds a lot of appeal, especially for those seeking clean desk setups. I’m using an MBP hooked up to dual monitors through a dock right now and it’s fine, but I do miss how clean my desk was back when it was a VESA mounted iMac Pro and companion monitor.
24" is a good size for casual users which is the main market. You can have two roomy-ish GUI apps (two web pages, etc) side by side.
This has been reported for many months. Sorry if you missed the news, but if you need a much higher memory variant the M5 Pro/Max are what’s available this cycle.
0: https://www.ifixit.com/Teardown/iMac+Intel+27-Inch+EMC+2309+...
1: https://www.ifixit.com/Teardown/iMac+M3+24-Inch+Teardown:+Ap...
https://www.apple.com/shop/product/mxk53am/a/magic-mouse-usb...
It’s nice for games but I don’t really care for productivity use. In fact in that situation I’d take integer-scaled HiDPI density over refresh rate in that scenario.
The ECC in DDR5 is on-die ECC and that's not "real" ECC, it's what DDR5 needs to not fail. Side-band ECC is the standard method, employing extra memory chips and 72bit memory. OEMs mix these in the marketing materials to fool the users into thinking they buy the better option.