I Want This $4000 AI Box 🔥

So, I don't know what this mini superco computer does, but I just know it's for AI, and I guess it's for local AI, and I kind of want to get it. Can you explain what it does for four grand? You sure want to get it? Yeah, it was three grand, by the way, when they announced it. Uh, but still, I Yes, if I had the money, I would. Look how cool it is. It's so tiny and it's so powerful for how tiny it is. I'm not so sure about this gold this gold pebble. Like it's like a stone wall. I'm not so sure about that. But so um so I'm gonna be totally frank like this only fell on my radar two three weeks ago something like that maybe a little bit more. Um and this is a tiny uh consumerra PC announced by Nvidia because so I I and again I don't know the history of Nvidia. Do they make PCs normally? There's a custom OS. You can see like actually this is running on a Mac. So I' I just I just don't even know anymore. Is this a Mac? It's that's running on a Mac. Yeah, it seems like it. Yeah. So is it Yeah. See, this is this is where my limited com knowledge comes into comes into play. So Nvidia basically have released this tiny little computer. Um 128 gigs of unified memory. It's basically built specifically for AI, right? Local AI. And it is quite pricey. Like four grand is quite a lot of money. But but what can it run? This is like the main question. Like can it run Haiku 4.5? Absolutely. Like this is this is um I watched a review of it. Um like a genuine review by I'm going to get him up because he's only I say he's only a small channel. who's bigger than us, but like uh you know, shout out to him. He's doing some great work. I'm going to Yeah. So, he was he was running comfy UI, which is for generating video. It's like a nodebased thing that he was he was running on it. And um it was doing fine. Like it was doing really really I want to see the LLM usage because I think he was getting into the I want to see his token usage because again, here we go. you can just see it live. He's got another he's got another comp competitive product on the on the left here. Um, so it's I think it's it's marginally faster this one. The G is it GeForce or GMX, whatever it's called? I don't know. So, you can see the results here. Like, yeah, I don't know. I don't know what to make of this speed because I think we're used to, you know, obviously hitting ChachiBT and it just just flying out. Like, are we supposed to be that impressed with this? Is this is this really is this really worth like you know locally running your AI like 4.67 tokens a second. So that's to me that's not that impressive. Yeah. But from a local LLM perspective like I you know people chime in where they want to. Like I think this is for the size that it is as well for the amount of wattage that it consumes. I think this is this is this is fine you know. Yeah, this is pretty cool. Can you chain them? I don't know actually. Like run them in parallel. That um that would be cool. Well um Oh, he's getting 64 here. Look, he's getting 64 with um OSS. That's um yeah, GBT OSS. So, I didn't I am I should have skipped through to his honest thoughts here, but um yeah, go watch that video if you if you want a bit more of a a genuine deep dive into it. Yeah, I I kind of want one, too. I don't know whether I want this one, but I you know, there's been times where I just kind of I would love just a little box just to know that I'm not hitting any server and that I'm just I'm coding completely off the grid. Do you know what I mean? Off offline. There's something in my gut that tells me that one day we will wish we had bought something like this sooner rather than later because Chachib goes down. All these things start like degraded service. We don't get they're not pumping as much money or whatever. So people resorting to buying machines like this and then but there's a shortage for some reason and it's like [ __ ] we should have we should have bought them when they were readily available. Bowen here. He walked into Best Buy to buy it, right? He was surprised. He thought there was going to be cues around the wall, around the corner or whatever. He was able just to walk in, buy it, walk out, all done. That That's fun. Yeah. Maybe maybe we're maybe we're living in in times where that's even possible. But, you know, maybe it won't be too long before something happens and we're wishing we had walked into Best Buy and bought it when it was freely available. Yeah. But for a grand, I I kind of want to have it because it's so cute and I I like the hardware, but I don't know if I could use something that slow. M5 just dropped. Apple just dropped a new MacBook with a just base M5 and they're advertising it for precisely that for ondevice AI. So, what I would love to see is like a comparison to see yeah, how big of a difference that would make. Maybe not necessarily with the base M5. Maybe something like M5 Pro or like M5 Max when they drop that. I don't know. Maybe later this month. But that would be really cool because if I could run um like a model on my just my on my Mac without any extra extra hardware that would be really cool. Let's let's not buy the idea that these companies are trying to sell us. Just look at the tech and how impressive it is that we have a model a mathematical model of all human knowledge packed in a machine like this. And if you could go back in time and you could bring this with you and somehow, you know, be able to turn it on with how crazy it would be, you know, have all that knowledge. I'm running Alama right now, right? Which is which is this is completely offline. I just asked it to write a 500word story. Let's go. Let's go. Um uh write uh sexy sequel. um 3,000 words and this is local on this is completely local, right? This is locally. So, it's doing um it's doing some thinking here. Is it an M1 machine that you have? This is M1. This is M1 Max, right? So, M1 Max. And what software is that that you're using? This is Lama. Look at it go. Oh, yeah. That that's quite fast. Yeah, that that's that's actually really good. And I could use it. It does make me want to try GLM. I think GLM is open source. I'm sure there's a way to pipe cursor into Olama, you know. Maybe I'd look into that and do a video on it. Can you like run cursor locally? Any like would you do that then locally, right? Yeah. Yeah. Yeah. Yeah. So any kind of AI request would be lama. So maybe I can play around with that and see if I can create. Yeah, that would be really cool because Yeah, we talk again we're talking about the Nvidia Spark here and you you know you want something more fast, but I've got the M1 Max here which is probably being outdone by the M5 already. Maybe not the I think the Pro I think it outdoes the Pro. Maybe not the Max, but the point is it won't be long before we've got some pretty um good enough. Yeah, I did have a look on Amazon, but I can't see it on Amazon or anything like that. So, I think we're still [ __ ] out of luck when it comes to the UK. I think it's just a US thing. But yeah, Nvidia Spark, pretty cool. Pretty cool. Nvidia Spark. Uh, I'm excited about it. 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.