Don't Buy an M5 MacBook Until You Watch This

People are loving their M1 chips. Like there are some people in here who are who are still rocking the Intel. I don't know whether I can get on board with the Intel uh bandwagon. And I think this is a really really crucial point like if you there's always this desire to like get the latest and greatest. But if you are not encountering any sticking points, then you you don't need to upgrade. Realistically, I know it's nicer to have the fastest, you know, and the latest and greatest, but you really, really don't. And this is the sentiment. Here we go. i5 chip. There we go. Matt Mini i5. Yeah. Um, and I think that's great. And I'm still very happy with my M1 um machine. It's still totally fine. And to reiterate the video, I've I upgraded to an M1, like top top of the line, which would have cost five grand or four grand if I bought it brand new, but I bought it off a guy down the pub. The point is is it I'm not even capturing that, you know, that um the full potential of it doing what we do. However, however, I have been doing some research and more specifically in the nature of the show, I've been looking into LLMs and running LLMs locally and how they work. Basically, it all comes back to unified RAM. Yeah. And the idea, you have your graphics card and that has a level of RAM in it. you know, you you'll get 16, you'll get 92, you'll get 128 gig in that uh in that RAM of that machine. And then you have your normal DDR5, DDR6 RAM chips, which you know plug into the CPU or whatever. Don't know the specifics around what it plugs into or whatever. Macs have this thing called unified RAM. So, right off the bat, you already know how much VRAM is available to your your graphics card just by buying RAM. Yeah. So, you said something along the lines of 16 gigs is the maximum you should go to. When we start talking about LLMs, which again, my my video went out today. I I run cursor using pointed to a local LLM and run local local um AI to power my agentic mode in inside of cursor, which was really fun. You you need to be very mindful that you need to load in the AI. Obviously, you need the space on your machine. uh let's say for example it's a 20 billion parameter model that takes up you know 17 gig of space or something like that that whole 17 gig needs to be able to load into your VRAMm of your graphics card and if you don't have that you can't run the model it just starts chugging and it just it just fails miserably so if we were to talk about AI if you were to look forward into local AI which I I'm getting more and more convinced by the day that local LLMs will become so so important. I going back to what I said previously, I think it's preposterous that we are controlling our browser with remote LLMs. I think there's so much that we can do on our local machine. We just need the machines to get faster and the the technology just to get a little bit better as with everything, of course. Yeah. But I still think we're the the graphics will be a bottleneck in all of that. So, what am I saying here? If you are looking into get into AI, local AI, and you are a designer or you are a developer, yes, the M1 will take you very, very far. The M1 Pro is the top spec you should probably go for um in all of that, not the Max, fine, but it will really struggle if you're running AI LLM and stuff like that. And that's maybe when you should start looking into the more recent models because I reckon I I kind of want to buy the M5 to be honest just to try it and see how much better it is. Um still need lots of VRAM that's the thing but the chip might ex that that will load the model but the speed at which it can um do the AI that you can have a a more reasonable chip like the M5. I just want to kind of try it. But yeah, it was sort of like as I was starting to understand more about Max AI, what these machines are capable of. I think what we were saying is good if you're strictly development, like old school development using remote AIs and design. That was a long speech, but I just thought I'd share what I've been learning and and that kind of went into the video that I I put in I released today. Um, so all I'm hearing is um you want a lot of a lot more subscribers so we can both afford M5 Ultra so we can [laughter] run our local LMS. Yeah, that's what we need, guys. Exactly. Exactly. Yeah. But now, you know, saying all that, I was so happy with my M1 and you know, as we were saying last week and as everyone's been saying, the M1 is totally fine for what we do. as I start to get more interested in local LLMs and AI in general, like it's like I'm looking I'm waiting for that M1 Pro and I'm like and then I would think M5 M5, sorry. M5. Yeah. Um I would I would completely go completely against your advice. I would max out the RAM because you know I've got I've got 64 gig of unified memory on this machine which can load a fair few amount of models but the the cores the speed at which my GPU is running that I mean I've got 30 core GPU but it's just not powerful enough to actually you know um I need to learn the vocabulary to to talk about this but like the speed at which the AI can run that's where the M1's really really struggle. they I can load the model but can't run it fast basically. Um but yeah, it's been really cool just kind of going on this nerdy journey. So go and subscribe to me [laughter] and watch that video and give it a like and give it a you know all that sort of lot. This is part of a larger conversation on my show Command AI which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.