> 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).
Off topic but doesn't intel do something weird where some i3 chips can use ecc but not i5 or i7 until you get into server chips? I remember seeing that when plotting nas builds a couple years ago. Should have pounced on actually building it then....
Yes, I liked using them for NAS builds. Great for ZFS machines.
Intel is weird with their stuff like that, I wrote about how they introduced and then dropped an SHA instruction that could be used to optimize hashing (useful for things like block checksum hashes on storage servers) before: https://neosmart.net/blog/will-amds-ryzen-finally-bring-sha-...
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.
I have a 60hz monitor and a 144hz monitor and I simply can't tell the difference in normal usage. I can see a difference on something like http://testufo.com if I specifically look for it, and until I checked that I assumed the OS was just lying to me and the monitor was actually set to 60hz.
It's odd because I can tell right away if a game is set to 60Hz instead of 144Hz when first launched before I adjust the settings. The mouse cursor just feels laggy.
Does she use f.lux on her usual rig? That saved my eyes like 15 years ago whenever I found it and that was on the low dpi screens to boot. Make sure you use the setting to match it to your lighting conditions.
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.
> 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.
What is it with fabs where there is just no competition springing up outside the existing memory cartels? Is the knowledge of how to build and run a fab strictly controlled by these companies? Is all qualified talent just immediately poached and put behind NDA and noncompetes? It just makes no sense. Capitalism would suggest these conditions with so much unstable equilibrium in this market would see increase in supply to take advantage of the pricing. The fact that this isn't playing out as such suggests there are some very powerful thumbs on the scale of this "free" market.
It takes about a decade to bring one up and during EVERY wave of a booming memory demand supply crunch there has been a subsequent round of major bankruptcies from suppliers who decided to scale and had no demand when they finally got online.
Supply is not elastic and responds very much slower than demand. There's no conspiracy or corruption needed for this result.
The conspiracy is real, it's even been proven in court, but as things stand, it's also necessary because of the factors you describe. Investors will not touch the DRAM market with a 10-foot pole, so it has had to rely on self-funding. While there have been state subsidies too, price fixing has proven to be the most effective way to survive the extreme volatility and fund the massive capex necessary for each generational leap. They have gotten a bit more sophisticated about it since the court case 20 years ago, but as far as I can tell, it's basically the same as it's ever been. The lack of effective competition is a function of the market itself, not any tilted scales. If anything, tilting the scales on purpose with greater state involvement is the only way to change the dynamics, but few states have the capacity and competence to carry that out effectively.
Micron's stock price did go way up, after they stopped selling DRAM to the consumer market. It has been flat for several months though.
For comparison, SK Hynix stock is flat YoY and Samsung's stock did go up this year, but has come back down a bit (they're also much more diversified). Those are the only other companies outside of China that actually make DRAM at scale. There's also CXMT, which is making inroads, but its stock is not traded on any open markets, and the company is heavily backed by the Chinese government.
>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.
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.
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.
After arguing and back and forth with Claude (not straight away), it found a kv streaming fork of llama https://github.com/achyudh/llama.cpp and an interesting ascii only model Qwen3.8-27B-ASCII-Condensed-UD-IQ4_XS.gguf . It's running, i'll see how it goes.
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.
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.
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.
Yes it runs on all levels of hardware - with the latest Strata my 5090 and 64GB DDR5 gets 200-240 tok/sec decode, 6500 tok/sec prefill. Around $0.09/M tokens in power usage (have excess solar capacity).
Jealous, lol. But I did spend a grand total of $500 for my setup. The rest of the server was built purchased with the use case of running game servers.
I asked Claude to come up with 7 problems to solve that were not part of any benchmarks. Besides speed, I asked it to put together a scoring that would reflect models that perform well on the 7 problems and in a reasonable time-frame (RWS, the definition is in the artifact). I valued models that produced good answers in a reasonable time better than models that had very high throughput but kept spitting useless stuff. Overall, IIRC most of them still had a >30% chance of producing wrong/non-working replies.
I then tested Strata ( https://github.com/niko1221/strata ) and the results are not even close. You can basically ignore everything in that table and use Strata instead, you can use bigger/less-quantized models and have way better results. Basically any otherwise-decent model degrades noticeably at Q < 5, and KV cache at Q4 also contributes a lot.
Still, a 20$/mo Claude subscription is way cheaper/more efficient if you code 2-3h in the evening than running stuff locally, unless you really have hardware than can hold models that get most things right at the first try. You can iterate way faster and cover a lot more ground in 2-3h using a Claude subscription, or OpenRouter if you prefer to change models. You get faster replies that are 99% correct at the first try, whereas using the local model you'd be spending 30% of your time fixing things that don't work and prompting it to do better.
Nowadays I use the local model to delegate tasks that would burn a lot of tokens, like sifting through big data archives for indexing and such.
QFN Q4 K XL Unsloth is basically identical to the bf16 model. It’s KL divergence is very very small. It is competitive with the full model. Also I benchmarked iq3 xxs on Strata for 3 days against deepswe and it scored better than the published benchnarks for the model. If the gsq rco quant prettain tasks fit well it is a great model quant. My data shows it takes 20% more tokens basically than q4 xl. A4 xl is within noise on all my repeated task benchmarks over the public model serving and my own runs of q8. I published the deeswe results. You have to be very specific about model choice and know these models and their quirks.
This is not about if a frontier AI provider is cheaper. It is about having control over your data and use of AI.
> QFN Q4 K XL Unsloth is basically identical to the bf16 model.
Yes, but like original poster I have 16GB of VRAM and 64GB of RAM, I was referring to non-Strata benchmarks :)
I did quickly run a couple things on Strata but haven't gotten around yet to benchmark bigger models/customize them with Strata, but will do rather soon. So far Strata/Swift IQ2_XS performed really well compared to non-Strata/Qwen3.8 quantized variants.
> This is not about if a frontier AI provider is cheaper. It is about having control over your data and use of AI.
Yep, totally get it, and to be honest getting to run ablated models sometimes can also be useful.
> Also YMMV in “faster” — I have less than 5K into my home AI lab shenanigans and it’s pretty fast :)
My rig is... ~1.5k, maybe 3k if we count the CPU and GPU upgrade since 2019 :-D I really wanted to grab one of the latest mini-PCs (Halo, Spark, and other minor assemblers) for local setup, but to be honest I can't justify a 6k+ purchase for me and my wife alone. Perhaps in 3 years these mini-PCs will get to the ~2k ballpark (perhaps used), and although older by then, I might be able to run things 24/7 locally as well.
With 32 GB more of RAM you can run ISTA GSQ RCO IQ3 XXS fast. You can make it work now but it is tight. You can stream experts from disk with builds now. You will get an okay token rate. Try it out if you have some disk space. It is a very good local model. I benched it at that quant. Sonnet 5 class at coding benchmarks.
Get a couple or V100S GPU. Figure out cooling. They will run it /fast/ with the right engine. :)
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.
Prefill isnt a huge problem with an MoE, but ya, dense models are a non starter before the M5. You are still better off using DeepSeek than a local MoE, but locally running something gives you possibilities if you are on an airplane without internet, or if you want to jump guard rails, or if you want to run a home device server, etc…
SSD token caching (like oMLX does) goes a long way for prefill. Typical miss rates are under 5% in long coding sessions, which is mostly the output of a turn's tool use.
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 jest...but I'm disgusted by the price gouging going on. Glad my hardware is relatively recent and I had mostly maxed out already when I bought it a few years back. The same setup today would be closer to $7k with equivalent parts.
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…
Waiting is never a stupid decision because you can always save a good deal on apple refurbished. I got a mac on bestbuy refurbished a couple years ago shortly after it was released. It had like 4 cycles on the battery and I saved hundreds.
Yes the latest chip is better but the rumor is that the M7 has either more/better parts for LLM math OR that it has significiant changes to memory bandwidth too.
I guess this means I’ll be hanging onto my M1 Pro for another year and seeing what M7 has to offer.
This is the first time I’ve ever kept a computer this long and never really felt like it needed to be upgraded. I suppose this is a bit of a gift and a curse for Apple.
More on the curse side, tbh. Mostly because RAM and flash chips have exploded in price and fabs get now booked by NVIDIA et al where previously Apple was the biggest dog and could demand better conditions.
But well, they'll survive :) Their M-series was a stroke of luck considering AI.
Probably not. The ROI for local AI is pretty terrible at best it's like 20 years to see a cost advantage. Most enterprises are going to be fine using hosted models. The two main reasons to use local AI are 1. education and 2. if you cannot use hosted AI for your use case (safety guardrails, payment restrictions, etc.)
I'm running qwen 3.8 flash next with strata (q3 quant) on laptop with 5090 and 128G RAM and OMP harness. At 90t/s speed it is pretty much good enough already.
Running Gemma 4 26B A4B and Qwen 3.8 27B on a M4 Pro with 24 GB - Qwen 3.8 runs fine but is slow and taxes the machine a bit, Gemma 4 on the other hand, is a better experience.
It seems so wasteful to spend premium on the hardware for local ai to be shoved into macbook form though. You'd probably save a lot building a compute node you shove under your desk that you connect to with some more comfortable but cheap thin client, maybe a refurbished low spec macbook pro (inb4 I say air or neo and people say they cannot work on inferior screens).
Then again the people working on stuff involving this much local model compute demands probably make enough where they don't care about the couple grand they'd save building their own node. I guess apple will get the bag from being the nicest prebuilt out there in this case.
Mini PC with Occulink, EGPU dock, 750w power supply and a refurbished 3090 is 2500. It can run Gemma4:31b at 1000+ tok/sec prefill, ~40 tok/sec generation (which is slow, but a fully connected model is way better for all tasks)
if you want more specialized partial activation models like Qwen, you can get 100+ tok/sec generation.
This beats anything that Apple has by a long shot. You don't need to run large models for usefulness. Large models can get things write the first time, but thats their only advantage. Running small models that are good enough like gemma4:31b fast allows you to do faster iterations.
And if you really want, you can also do multi GPU setups if you want slightly slower model loading with usb4/thunderbolt ports.
TLDR; local ai has already been useful for quite some time if you arent brainwashed into believing that Macs are the best for running local inference.
> "Man you Apple Fans really live under a rock... ...refurbished 3090 is 2500"
Before you make assumptions, there are valid reasons for the choice I made:
I paid $1k for a refurbished M1 Max 64GB macbook that can run Qwen 3.8 27B at 24tok/sec (looks like they are about $1.5k on ebay). In my case, Apple was a lot cheaper (used) if you're looking at price-per-token. Yes, it's almost twice as slow as a 3090, but it's also half the price /performance ratio.
Qwen 3.8 27B Q4 with 64k:
3090: ~40tok/sec
M1 Max : ~24tok/sec (MLX)
In fact, the Max can even do 265k context which isn't possible with a single 3090 last I checked (without dropping to Q3-Q2)
Go ahead and run Gemma4 on your M1, and you will find that its unusable at 12 tok/sec.
And this is also not touching prefill, which is the biggest thing that you need fast, and Apple sucks in this regard compared to nvidia cards. With gfx cards, you can throw big codebases at the local models and it will happily crunch through them in very little time.
Sure, spending 1.5k on a Macbook enables you to run a local model, and if you really have a good reason for running local (like running heretic/abliterated models) for tasks you don't mind waiting for, it works. The difference is, spend 1k more and let you never have to pay for tokens, and you can run inference all day long without your computer thermal throttling, like your Macbook.
And I don't need to make assumptions when you say idiotic stuff like "We're getting close to local AI being really useful", like you have any idea what you are talking about. On par with being an Apple fan though.
A $2500 Mac Studio with M5 Max running Qwen3.8-27b or Gemma4:31b in oMLX gets about the same speeds you quoted... and it comes with a warranty and has enough memory not to choke at longer context lengths.
Srouji is killing it on whatever he has on mind or is asked to do. Everyone is riling up on GPU and Cloud AI costs. 99.99% of what people use AI for today already processes fine on an M1 from 5 years ago.
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.
Intel macs are fine for non power use. We still have a 2012 that does that. Does chrome spin the fans and is battery life shit? Yes. But it doesn't struggle with it at all. I mean its 4 2.0ghz cores and 16gb of ram, ssd. The other 2012 intel mac I had was dual core but a big dual core, 2.5ghz, and I had that with 16gb and ssd. My m3 pro is only 2.4ghz per core. Graphics is a lot better than intel hd but you aren't going to deal with that in non power use case. m3 pro has all this power and I have to use wine to run basically every game. TF2 over wine on this thing has actually slightly worse performance than TF2 on windows on those intel macs using bootcamp. Native ports are great, civ 5 is basically instant and cold running while barely running on the intel macs, but they don't make many native ports anymore. Old studios that made mac native ports no longer do them. Valve doesn't anymore. Paradox doesn't anymore. AAAs don't really do it anymore either.
The modern iMac is the dream desktop that I'm sure Jobs would have loved to live to see, but Apple's laptops have gotten so good there's practically no reason to buy a computer from them that you can't take on the go.
It only makes sense for niche use cases, like your grandpa who is happy to have a single spot in his home for the computer, or a reception desk at a business, etc.
This is basically 99% of the proposition of the iMac at this point. If you don't need the specific all-in-one, color, and other design aspects of the iMac or can afford a Studio display, there's no reason not to buy some other display + a Mac mini or a laptop + display
I actually think a desktop makes sense if you want to run local LLMs. I have an M4 Max laptop left to me when I got laid off. It can run good-sized LLMs, but the system heats up, the fans blast off, and it rips through the battery in under an hour. I'd rather have a desktop set up as a private inference server that I could connect to remotely.
Bingo, desktops are still king for actual raw processing. Apple laptops are great, but they're still just computers, given proper cooling and power they can perform even better.
An imac doesn't strike me as a desktop with proper cooling. Maybe they do some funny business in the chassis, but its pretty clear that airflow has been put on the bench in favor of having a sexy thin backside.
Desktop maybe, but Imac that costs more for the same specs of the mini you can shove in the drawer in a desk? I mean with the airspace taken up by an imac you only remotely connect to you could probably have two side by side stacks of I don't know, maybe 20 mac minis each? HPC territory.
But the iMacs only have the sans-adjective M-series processors. And is it really gonna have optimal thermals in that thin frame pressed against a big, bright high-res screen?
If you need to push performance you'd probably go for a Mac Studio, no? Or at least a Mini.
The new post Jony Ive design and limitation to 24 inches make it undesirable. I know a lot of folks with older 2007 - 2019 iMacs but none of them who upgraded from those have the new 2020+ one.
I think it's mainly that from 2006 to 2019, the iMac was the only mid-tier desktop that Apple sold. Today, you can get a Mac mini or MBP that's the same or faster than an iMac.
There was a quad core Mini back in 2011. The old iMac was a good all in one package for designers, especially because you'd spend half or more for an equivalent quality screen.
I did have a 2010 Mac mini server, don't remember how it compared to the non-server one other than having dual HDDs instead of the DVD drive. I put them in RAID-1.
I agree with you, but Apple's monitor situation is letting everybody down. Not having a 32" monitor for professionals is frustrating. I am glad I bought my Pro Display XDR.
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.
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).
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.
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.
Intel is weird with their stuff like that, I wrote about how they introduced and then dropped an SHA instruction that could be used to optimize hashing (useful for things like block checksum hashes on storage servers) before: https://neosmart.net/blog/will-amds-ryzen-finally-bring-sha-...
Her eyes were much less tired at the end of the day.
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 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.
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.
The fabs are terrified to overscale in case the market collapses before the builds are complete.
> AI datacenter provisioning will go down
Depends on political opposition at both local and national levels. Hard to predict the future, which feeds into the previous point.
Supply is not elastic and responds very much slower than demand. There's no conspiracy or corruption needed for this result.
Have you not seen Micron or Sandisk stock?
Micron's stock price did go way up, after they stopped selling DRAM to the consumer market. It has been flat for several months though.
For comparison, SK Hynix stock is flat YoY and Samsung's stock did go up this year, but has come back down a bit (they're also much more diversified). Those are the only other companies outside of China that actually make DRAM at scale. There's also CXMT, which is making inroads, but its stock is not traded on any open markets, and the company is heavily backed by the Chinese government.
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.
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.
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.
Are you sure or is this new?
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 ($500 GPU).
I get around 40-60 tok/s depending on the quantization level I choose.
As you can see, the CPU is nothing special, don’t even need DDR5.
I asked Claude to come up with 7 problems to solve that were not part of any benchmarks. Besides speed, I asked it to put together a scoring that would reflect models that perform well on the 7 problems and in a reasonable time-frame (RWS, the definition is in the artifact). I valued models that produced good answers in a reasonable time better than models that had very high throughput but kept spitting useless stuff. Overall, IIRC most of them still had a >30% chance of producing wrong/non-working replies.
I then tested Strata ( https://github.com/niko1221/strata ) and the results are not even close. You can basically ignore everything in that table and use Strata instead, you can use bigger/less-quantized models and have way better results. Basically any otherwise-decent model degrades noticeably at Q < 5, and KV cache at Q4 also contributes a lot.
Still, a 20$/mo Claude subscription is way cheaper/more efficient if you code 2-3h in the evening than running stuff locally, unless you really have hardware than can hold models that get most things right at the first try. You can iterate way faster and cover a lot more ground in 2-3h using a Claude subscription, or OpenRouter if you prefer to change models. You get faster replies that are 99% correct at the first try, whereas using the local model you'd be spending 30% of your time fixing things that don't work and prompting it to do better.
Nowadays I use the local model to delegate tasks that would burn a lot of tokens, like sifting through big data archives for indexing and such.
This is not about if a frontier AI provider is cheaper. It is about having control over your data and use of AI.
https://stealthgoat.com/fnfl/pelicans.html
Also YMMV in “faster” — I have less than 5K into my home AI lab shenanigans and it’s pretty fast :)
Yes, but like original poster I have 16GB of VRAM and 64GB of RAM, I was referring to non-Strata benchmarks :)
I did quickly run a couple things on Strata but haven't gotten around yet to benchmark bigger models/customize them with Strata, but will do rather soon. So far Strata/Swift IQ2_XS performed really well compared to non-Strata/Qwen3.8 quantized variants.
> This is not about if a frontier AI provider is cheaper. It is about having control over your data and use of AI.
Yep, totally get it, and to be honest getting to run ablated models sometimes can also be useful.
> Also YMMV in “faster” — I have less than 5K into my home AI lab shenanigans and it’s pretty fast :)
My rig is... ~1.5k, maybe 3k if we count the CPU and GPU upgrade since 2019 :-D I really wanted to grab one of the latest mini-PCs (Halo, Spark, and other minor assemblers) for local setup, but to be honest I can't justify a 6k+ purchase for me and my wife alone. Perhaps in 3 years these mini-PCs will get to the ~2k ballpark (perhaps used), and although older by then, I might be able to run things 24/7 locally as well.
Get a couple or V100S GPU. Figure out cooling. They will run it /fast/ with the right engine. :)
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.
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 jest...but I'm disgusted by the price gouging going on. Glad my hardware is relatively recent and I had mostly maxed out already when I bought it a few years back. The same setup today would be closer to $7k with equivalent parts.
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.
Yes the latest chip is better but the rumor is that the M7 has either more/better parts for LLM math OR that it has significiant changes to memory bandwidth too.
This is the first time I’ve ever kept a computer this long and never really felt like it needed to be upgraded. I suppose this is a bit of a gift and a curse for Apple.
But well, they'll survive :) Their M-series was a stroke of luck considering AI.
Then again the people working on stuff involving this much local model compute demands probably make enough where they don't care about the couple grand they'd save building their own node. I guess apple will get the bag from being the nicest prebuilt out there in this case.
Mini PC with Occulink, EGPU dock, 750w power supply and a refurbished 3090 is 2500. It can run Gemma4:31b at 1000+ tok/sec prefill, ~40 tok/sec generation (which is slow, but a fully connected model is way better for all tasks)
if you want more specialized partial activation models like Qwen, you can get 100+ tok/sec generation.
This beats anything that Apple has by a long shot. You don't need to run large models for usefulness. Large models can get things write the first time, but thats their only advantage. Running small models that are good enough like gemma4:31b fast allows you to do faster iterations.
And if you really want, you can also do multi GPU setups if you want slightly slower model loading with usb4/thunderbolt ports.
TLDR; local ai has already been useful for quite some time if you arent brainwashed into believing that Macs are the best for running local inference.
Before you make assumptions, there are valid reasons for the choice I made:
I paid $1k for a refurbished M1 Max 64GB macbook that can run Qwen 3.8 27B at 24tok/sec (looks like they are about $1.5k on ebay). In my case, Apple was a lot cheaper (used) if you're looking at price-per-token. Yes, it's almost twice as slow as a 3090, but it's also half the price /performance ratio.
Qwen 3.8 27B Q4 with 64k:
3090: ~40tok/sec
M1 Max : ~24tok/sec (MLX)
In fact, the Max can even do 265k context which isn't possible with a single 3090 last I checked (without dropping to Q3-Q2)
Try more like 70-130 tok/sec. https://github.com/L-Forster/open-jet/tree/master/megakernel.
And also Qwen models famously chase benchmarks. They are good for writing code, but fall behind in other areas. Gemma4:31B is 6gb larger and does a much better job across all areas. https://www.reddit.com/r/LocalLLaMA/comments/1vyzopv/gemma4_...
Go ahead and run Gemma4 on your M1, and you will find that its unusable at 12 tok/sec.
And this is also not touching prefill, which is the biggest thing that you need fast, and Apple sucks in this regard compared to nvidia cards. With gfx cards, you can throw big codebases at the local models and it will happily crunch through them in very little time.
Sure, spending 1.5k on a Macbook enables you to run a local model, and if you really have a good reason for running local (like running heretic/abliterated models) for tasks you don't mind waiting for, it works. The difference is, spend 1k more and let you never have to pay for tokens, and you can run inference all day long without your computer thermal throttling, like your Macbook.
And I don't need to make assumptions when you say idiotic stuff like "We're getting close to local AI being really useful", like you have any idea what you are talking about. On par with being an Apple fan though.
I'm one of those Mac users, but I do have a linux tower for remote dev work and LLMs.
Or at least give us the option for a black display surround. The white bezel around the display is such an eyesore...
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.
It only makes sense for niche use cases, like your grandpa who is happy to have a single spot in his home for the computer, or a reception desk at a business, etc.
This is basically 99% of the proposition of the iMac at this point. If you don't need the specific all-in-one, color, and other design aspects of the iMac or can afford a Studio display, there's no reason not to buy some other display + a Mac mini or a laptop + display
If you need to push performance you'd probably go for a Mac Studio, no? Or at least a Mini.
I did have a 2010 Mac mini server, don't remember how it compared to the non-server one other than having dual HDDs instead of the DVD drive. I put them in RAID-1.
If you need portability, you have your phone already.