16:11
A few years ago, when the M2based Mac Studios first came out, I started to switch my shooting setup over to something that I don't think many people do in this space, I started to try to edit my videos in real time, like during my shoot. So, I shoot off of a few cameras, but this one right here is like the main camera I shoot from. This is like we call it the A cam in the business, but this is like the facial shot. And then, uh there's this like table shot. We have an overhead camera and then we also have this like back camera that uh we use for like different angles and all four angles are visible here and this is all real time right like this is something that I try to capture um all at once and all the edits
like switching between the different angles like the face the overhead the table like this is all done with an Elgato I'll switch between them and the purpose of it is just to make it as fast as possible for me I try to make this stuff in as seamless and as like organic panic of ways possible. I don't use teleprompterss and scripts on anything. I just do this. I talk about the thing that I want to off my dome and I just like try to do the best job that I can in as little time as possible. Now, the reason why I do it all off or I used to do it all off of the M2based Mac Studio, it's not even this one. This is the newer one, but back then I did it off the M2 Ultra was because all of these
cameras, 4K cameras being streamed and encoded in real time while doing all the transitions could be done silently while I was doing this. Like I hate fan noise when I'm doing this. It's something that's like just a me thing, but I like it to be really quiet when I shoot these things so I can concentrate on the delivery of the message that I'm trying to share with you guys. Now, a couple years ago when the M4 Mac Mini came out, I tested out an M4 Pro just to see if it could keep up with this workflow. And it could, but the GPU on that system, the M4 Pro, was taxed quite heavily, like 75 80% usage. And during the summer when it was a little bit hotter, sometimes the fan would kick in and there would be fan
noise and it was like, okay, I'm just going to stick with the Mac Studios. But there's a new Mac Mini. If you're wondering why this is here, this is because I can control the back camera with the stick. Sometimes I want to move the camera around. But there's a new Mac Mini with new CPU options, much faster CPU and GPU obviously. So, you're going to get big gains in a lot of applications and workflows. But on my particular setup for this recording workflow, I'm noticing way better GPU pressure. So, this is now sitting at like maybe 60%. A little bit less and it's going full blast. So, the past few videos I've done on this system, the Steam Frame review I did on that, same with the iPhone review, and it's because
this is such a quiet, it's so good. It's replaces the Mac Studio for everything that I use it for. This particular system is running the M5 Pro with 24 gigs of RAM, and for what I'm doing, I don't need more RAM than that. You can also run local AI models on it. So, the M5 Pro chip has higher memory bandwidth than the M6 chip, which makes models run a little bit faster. And you can get more memory on the M5 Pro configuration, so you can run bigger models if you want. Now, in terms of pricing, this generation starts at the M6 chip. You get 16 gigs of RAM, goes up to 32, and it starts at $900. It is a significant bump up in price than the previous generation. Uh the higher tier configuration with the M5 Pro chip maxes
out at 64 gigs and it's a substantial step up in performance if you can take advantage of the GPU or multi-threaded CPU performance or if you're into AI workflows. Now AI workflows is what made that M4based Mac Mini wildly popular. So, when it came out, it was like crazy because this was a tiny machine that was energy efficient, silent, and very inexpensive for the AI performance that you're getting out of it. So, enthusiasts were buying these things just because they like to play with local models, but you could cluster these things. You could get an army of Mac Minis, connect them up, and they would work together with the combined RAM of the full stack. So, you could have very large models serving a whole
team of users, all done locally for a relatively inexpensive price. But the connection between those M4-based Mac Mini nodes weren't super fast. So, the cluster couldn't perform at like ultra high speeds. It just had a ton of memory to work with. Which brings me to the new M5 Ultrabased Mac Studio. So, this is, you know what, I can't tell which is one, but one of these is the M5 Ultra. The other is the M3 Ultra. Now, the one I had the previous generation, the M3 Ultra, had 512 gigs of RAM. The new M5 Ultra that I've been sent only has 256 gigs of memory. I obviously have to send these back. I wish I could keep these things, but the performance on the M5 Ultra device is so ridiculously fast. So,
every time they come up with one of these Ultra chips, I'm always blown away by just how much faster it is than the previous generation. the GPU, the CPU, but particularly the memory bandwidth this time is just off the charts. It's just so fast. It's obviously going to be the fastest Mac hardware you can get. But even compared to Windows stuff like the 5090, this is really good. I tested out Draw Things, which does local generative AI video. I personally don't love the idea of this. I do understand that certain industries lean on products like this, like I don't know, if you do like mockups or, you know, storyboarding and you want something quick and inexpensive, a tool like this is great.
But when I look at it, I'm like, I feel icky even just using a benchmark in it. But it is interesting that this hardware can do it as well as it does now. It's also super fast with LLM because the memory bandwidth is crazy high. And there will be a 512 GB configuration of the M5 Ultra coming in the near future for the people that have the need and more importantly the budget to get that thing. And lastly, there's clustering. So Thunderbolt 5 supports the RDMA connection that I was talking about before. So you can get two Mac Studios even of different generation. Uh the M3 Ultra with its 512 GB combined with the M5 Ultra with its 256 GB. You now have a combined memory of 768 I believe. And
you can run this in a single pool. So you can now hold models that neither one of these could have held individually. But keep in mind it doesn't make them run faster, right? It just allows the model to be larger or you're going to have longer context for whatever it is that you're doing, but it doesn't inherently make it like it doesn't make the thing run faster just because there's more RAM. Uh maybe your workflow might be faster because of that extra RAM, but it doesn't make the model like it doesn't uh process it faster. Um okay, so there you have it. This is the new stuff from uh the Mac lineup. These things are obviously crazy expensive. I will say having done a little bit of more testing with this
stuff, it still holds true this like the whole thing about cloud being significantly more cost effective like you could get years of really high token monthly subscriptions for the cost of one of these machines. Uh obviously when you have something local something like a Mac Studio you have the massive benefit of like privacy and obviously there's certain like types of data that you are not even allowed legally to upload into a cloud. So there's tons of workflows that would depend on something like this, but for the average workflow, I think you're going to get with cloud services speed. It's going to be faster, better, stronger, cheaper, all of it.
It's just local does have some benefits. Okay, there you have it. The new Mac lineup.
