>I realize I’m somewhat limited (16GB RTX 9070), but still, it seems really far off from the kind of experience even a basic $20/month subscription gets me.
I just ordered a new Mac Studio M5 Max 128GB $5899 ($6400 with tax) to be able to run the bigger "consumer size" models in the 70B parameter range (~96 GB). That said, I have no illusions that this expensive setup with a Qwen Flash coding LLM will be comparable to a $20/month subscription. Even upgrading to an even more expensive Mac Ultra 256GB for $10000 to hold a bigger model still won't be comparable. Apple hasn't shipped my Mac yet and I'm still considering cancelling it and downgrading to a smaller 64GB RAM config ($4299) to save $1600.
Why did I initially spend the extra $1600 if I knew ahead of time that it wasn't as good as cloud AI? Because I thought I could use some local LLM for the easy tasks or when I hit cloud rate limits. No issues with privacy so that wasn't part of the motivation at all. I just wanted some local AI capability to augment a subscription. I've not totally convinced myself of the cost/benefit of this.
Based on today's consumer hardware landscape, you're paying very high prices for crippled capability compared to the cloud AI subscriptions. We're also in a transition period where the next iteration of hardware improvements have some compelling features for local AI. Apple's upcoming M7 (2027 or 2028) is anticipated to have better GPU and neural engine to help with prefill TTFT. AMD Strix Halo is about to release 192GB system which is a big upgrade to their current 128GB ai pc. Maybe apply my $1600 savings towards those newer products. Those future products will still be very expensive but maybe the cost/benefit will be better.
> Why did I initially spend the extra $1600 if I knew ahead of time that it wasn't as good as cloud AI? Because I thought I could use some local LLM for the easy tasks or when I hit cloud rate limits.
The maths don't check. With Deepseek Flash one goes a very long way with 1600$ - even 10$/month, for easy jobs, are more than 13 years, and at a higher quality.
Oh no doubt. But one does have the guarantee that no bits left their home and that's a big one for some.
The low hanging fruit stuff for me is more something I use it for because I have the local LLM setup running anyway. It wasn't the reason I bought it, but now that it's there I might just as well use it as much as I can.
I think the M5U Ultra 96gb is the sweetspot in that price range. It has more compute and bandwidth so you get to run models better sized to its hardware. I believe the Max would be too slow; personally I'm getting this SKU because I think it'd suck to get the 128gb Max and then discover it's too slow, and end up just using cloud providers anyway.
I agree about getting the Ultra if you're interested in AI (LLM) inference speed. I'm a little perplexed as to why there isn't a RAM option in between 96GB and 256GB, though. For instance, I believe Deepseek v4 flash runs a lot faster on (https://github.com/antirez/ds4) with 128GB than 96. I assume the higher SKUs have low enough sales that Apple doesn't want to be manufacturing too many different RAM configs as that would eat into their margins. (I say this as someone interested in AI hardware in general, even the 96GB is out of my price range).
It's the upsell ladder. This is how they get you to buy a more expensive tier or two. At first I was only going to look at 64gb or 128gb Max, but ended up here.
I also think it’s possible that it may be best to continue waiting on the Mac side of the house despite the increasing prices.
I think that some of the hardware design folks have been blindsided by AI demand and we haven’t really gotten that next generation AI hardware yet, to the point where buying M5 isn’t going to make sense in a couple of years.
Rumors seem to be that the M7 is the generation that Apple is looking to push AI performance much further.
I’m not sure that Apple anticipated this specific route that computer hardware has gone and I don’t think M5 and previous iterations were really specifically architected for local AI performance, more like they happened to be pretty good at it.
Local LLMs are improving for fixed hardware, though - a 30b parameter model now is markedly better on the same hardware than one from a year ago.
Keep the memory. You’ll be glad you did when you realize that you’re better off with a solid coding model plus a good voice model and also a lightweight all-rounder all running at once isn’t of loading dynamically (slowly). It also helps if you want to be able to run a browser, IDE, and container environment.
> AMD Strix Halo is about to release 192GB system which is a big upgrade to their current 128GB ai pc
Big upgrade to memory capacity but memory speed is only going up by a few percent, so its still going to be slow with more than a few B active params (I have one)
Serious question: why not run DGX Spark or Framework Desktop, at 30%-50% lower cost?
M5 Ultra has 4-5x the memory bandwidth of both. 1.2 TB/s memory bandwidth opens up good performance on relatively large models.
you can get 4xGB10 for <20K so that gets you about the same tg and pp will be probably better. Power consumption though will be something like 200W idle so that's a bummer. And you get VLLM and SGLANG unlike them mac where one has to use oMLX (nice but not the same concurrency or cache performance) I have 128GB M5 Max laptop and sill prefer to run things on other boxes in basement because its no fun to have the primary device being overloaded.
[dead]
It not only about it being an expensive setup (or not), and also other considerations:
- There's no guarantee of the $20/month service, and it likely has some limits compared to dedicated hardware token wise.
- Model are becoming more and more efficient, in many cases an M1 Max Mac Studio is still capable with 32 GB. 128 GB ram may not be the necessary baseline.
- Folks may think they want to only have a general model running locally (it's the comparable after all from the cloud providers), but we have to remember if the tasks we're trying to do ultimately are more specific than general and if there's space for the smaller models to do that.