I’ll spare you the unbox pron and the benchmark LARP… I mean a bit… and get right down to it. The M1 Ultra is pretty much the ultimate expression of Apple’s scalable silicon architecture, and delivers on that promise… almost linearly. And I’ll get to why “almost linearly” in an Intel TJMAX hot minute. Right after I pull a Linus and thank today’s sponsor, the CuriosityStream and Nebula bundle, which you should jump on faster than M1 Ultra because I’ve got a whole new, exclusive Studio Tour series already going up!
So, basically, if there’s anything you couldn’t do on an M1 Max MacBook Pro, if there was any limitations you were facing or walls you were hitting, you can break right through them, most of them, with the M1 Ultra in the Mac Studio. Those limitations, those walls, like… the Kool Aid man.
Because, where M1 Max had the exact same CPU and Neural Engines as the M1 Pro, albeit it with double the potential GPU, media encoders, RAM, and memory bandwidth, M1 Ultra doubles everything. Every core, every feature, everything down to the transistors. To a ridiculous degree. I mean, we now have two secure enclaves, which I can only guess is like having two Klingon hearts, and two iPhone 12-class image signal processors… in a box with no camera. But that’s the only quasi dark matter compromise in what’s otherwise a blindingly bright solution to making an already massively monolithic die, literally twice the size.
Because that’s what M1 Ultra is — two M1 Max dies joined together with a silicon interposer running 10,000 connections at 2.5TB per second, basically an OG server farm worth of glue, for double the CPU, or 4 icestorm efficiency cores running at 2GHz, and 16 firestorm performance cores running at 3.2GHz, spread across those two dies in two clusters each… giving you something we never could with Intel Xeon in previous pro-level Macs — massive multicore parallelism without sacrificing single core performance. Along with the ability to maintain that single core clock in one cluster even if another has to down clock some to fire all the cores. I’d say it’s like having a carpool lane on the highway, but because Apple’s performance controllers are so good at dispatching between the various cores to begin with, and 800 GB/s of memory bandwidth is something we only ever see on high end GPU, not CPU, so there’s such low latency, even across dies, that it’s virtually indistinguishable at the software level… traffic is so highly optimized it ends up being more like having a slipstream… alongside a hive fleet. So, if you had any workloads that were purely or mostly CPU constrained, even on M1 Max, you’re realistically throwing twice as many cores at them now.
Also twice the 4th generation Apple Neural engines, or ANE. There were 16 cores in A14, and 16 in every M1 through Max. But with double Max dies comes double ANE. Also, double the AMX accelerators on the CPU, and double the ML controllers for tying those together with the GPU for the CoreML framework, and making it all… double everything, basically. Just not in the same way as GPU… which I’ll get to in a second. Because, my understanding at least, is that these two ANE blocks are still treated as two separate ANE target by the system, not as a single, abstracted ANE target. Which means, single ML workloads will get dispatched to a single 16-core ANE block, resulting in the same serial performance as M1 Max. But, multiple ML workloads will get dispatched to both 16-core ANE blocks in parallel, resulting in… potentially up to twice the throughput, depending on the exact workloads. Especially in easing bottlenecks Whether that’s training ML models or doing AI image or video enhancement.
And, of course, twice the G13 GPU cores, up to 64 of them now, 32 on each die, but thanks to Apple’s metal frameworks and the ridiculous amounts of bandwidth involved, the two blocks ARE presented as a single GPU to the system. Whether that really matters to you in the real world, and how much you can take advantage of it, will depend a lot on software and workloads you’re running. What I mean is, where up to 128GB of RAM isn’t out of the ordinary at all for a Mac CPU, especially not when the iMac Pro maxed out at 256 GB and the current Xeon Mac Pro at 1.5 terabytes, having 128 GB of RAM available to the GPU is, I think, unprecedented. Like, 3D models and environments at a size and with polygon counts and effects that you can load up and manipulate in real time in a way that just isn’t possible in any other desktop box
The only major harsh in this mellow being, of course, that NVIDIA has CUDA cores, and ray-tracing, and other features embraced by high-end gaming studios, and similar if not the same for AMD, along with DirectX, and Vulkan, and Steam is making Proton work on the SteamDeck. Apple is kinda stuck in build Metal and they will… oh god, they will come, won’t they? We made snacks, why aren’t they coming? Even big games that get ported over to the iPhone and iPad — allow me to point you to any recent Dave2D video! — just aren’t often making it over to the Mac, even as check-the-box-to-let-the-iOS-app-run-on-Mac ports!
And since time might not be enough to fix that, if Apple really considers it important, they may have to flex their combined market share and, more importantly, money, to Kaiser Soze-single-act-of-will-it to happen.
Then there are the double media engines, many of which were already doubled from M1 Pro to M1 Max, so are… double double’d in M1 Ultra. Like your typical Tim Hortons order. Sorry, Canadian humor. So sorry. But that means twice the decode and encode accelerators for H.264, HEVC, and critical to higher-end video work, ProRes. Up to 18 simultaneous streams of 8K ProRes, to put Apple’s number on it. So, for example, if you were making Fullest House for Netflix and you wanted to show a 6 x 3 grid of Uncle Jessie and the grand-niece hijinx scenes from the pilot you could edit and play it all back in real time. Which is… a real time saver.
And I know a lot of you audio techs and developers and game designers get super salty when every video you watch focuses almost completely on… video. But, one, that’s not surprising from people who’s literal job is to make videos. Two, Apple is clearly focusing on video workloads with things like the dedicated media engines. And three, video production is one of the very few things that actually lights up a ton of silicon. With those media engines handling encode/decode, GPU handling effects and accelerating presentation, and CPU sometimes acting like a parent for metadata management, but also pitching in with the gnarlier camera codecs not accelerated by the media engines. It’s… a perfect storm of pain for a modern compute system, and one that Apple is actually turning into more than a bit of pleasure with their custom silicon.
You also get double the I/O controllers, three USB and three thunderbolt. That gives you four Thunderbolt 4 controllers, 10 Gb/s ethernet, 2 USB-A, and HMDI 2.0 on the back. No MagSafe like the iMac, alas, because no external power brick, like the iMac, which MagSafe would require. On the front you get two USB-C or two Thunderbolt ports, and an SDXC card reader.
Though this is one of the few areas where I don’t think we’re seeing real linear scalability. From M1 Max to M1 Ultra, the two USB-C ports on the front of Mac Studio do benefit by becoming full-on Thunderbolt. So you can drive a display right up front if you want to. But the SDXC card reader on the front is still UHS-II and HDMI port on the back is still 2.0. No bump up to SDUC and UHS-II, never mind the kind of CFExpress cards I’m using these days, or to HDMI 2.1… which is an absolute joke of a mess of a standard at this point, but would allow for even more bandwidth for even higher resolutions and frame rates, among other amenities. It could just be that M1 planning was long enough ago that those standards lacked the popularity and maturity to be considered for Apple silicon round one, but it does legit suck for future proofing. Either way, if you want them, you gotta dongle up a type C port to get them, but as any working pro knows, standards evolve fast enough we never really get to live an adapter free life anyway.
Now, M1 itself was transformative. It focuses on performance through efficiency, meaning this no longer so little grown up from iPhone and iPad chip runs ludicrously cooler than Intel and AMD’s shrunk-down from enterprise server chips, meaning it gets profoundly better battery life in the same enclosures, and can enable enclosures that simply weren’t thermally possible before.
It has better performance than many chips that consume way, way more wattage, meaning it gets workloads done faster, saving you dozens of minutes out of every hour, every day. Like I hit render on a video, go to get a coffee, and it’s done so fast I think I forgot to hit render, and then forget to get my coffee… Which, win or loss, I honestly couldn’t tell you. No clue.
It’s as immediately responsive as an iPad, which sounds silly, until you use it, and you realize how many dozens of seconds of every minute you wasted waiting for Intel to stop beach balling when you were just dragging a effect a few frame one way or tother. Or, you know, down arrowing through files in QuickView, without having to wait an excruciating amount of seconds for every single image or video to render, like it did with Intel more-like SlowView. Everything is just instant, which is such a massive quality of life improvement, even 18 months in, it still makes me smile on the daily.
With M1 Pro and Max, it was also like basically getting a computer and rendering box two-in-one. Where, previously, you’d export a video and it would just grind the Intel CPU to a halt, making Thumbnail work tough, even just web browsing a hassle, now that’s all just handled by the media engines, including ProRes, leaving the CPU completely free for any all other non-media work. In that specific situation, it’s legit like having a second Mac. So, you can work in parallel in a way that just wasn’t realistic, as in, super frustrating, before.
And now, with M1 Ultra, you have all of that just doubled again. In a way that… kinda makes me wish Apple wouldn’t stress performance so much. Because it’s not really a performance story. Marketing aside, anyone, everyone, would tell you Apple silicon can be beat on performance if a competitor is just willing to spend the voltage. Just jack up power to win that benchmark LARP crown from the download, run-and-done crowd. Which everyone from Intel to Qualcomm are clearly more than willing to do. But it’s really an efficiency story. Not just how efficient Apple silicon is, or that it achieves its performance through that efficiency, but how efficient it allows us to be, how performant it lets us be through that efficiency as well. I as a person am not faster with Apple silicon, but the amount of parallelism, integration, off-core features, and overall design lets me get way more done than I ever otherwise could. It’s not a gas guzzling roadster, it’s hybrid SUV that lets you move so much more, so much better, and in a way that makes me even more curious what the multi-die scalability limits will really prove to be
And lest you think I’m treating the new Mac Studio as merely a bead blasted aluminum candy shell for M1 Max and Ultra, I kinda really dig the industrial design and whole concept.
It’s like, you’re Apple and this lightbulb goes off in your head. All Incandescent. Neon even. There are pros who want the most powerful Mac imaginable, but don’t care one iota about internal expansion slots. So, you make the Vader helmet meets trash can 2013 Mac Pro. But you overestimate the impact of OpenCL, and underestimate the saltiness of pros for whom Mac Pro doesn’t just mean power but also means modularity and expandability. So, you Craig-a-culpa the thermal corner, resurrect the cheese grater in the 2019 Mac Pro, and you give those salty prose the name and the Mac that really means something to them. But that lightbulb, it’s still there. Like an itch in the brain. And you’ve already shipped a higher-end Intel Mac mini. The original, the classic, all-in-none. And that lightbulb just goes full on floodlight. You don’t need a Mac Pro mini, you need a Mac mini Pro. Screw that, Max. Then you put it on, like Hulk serum, and make it full-on Ultra. And then you give it a 27-inch Studio Display, which I’ll have a whole video on, so don’t forget to hit that subscribe button, and you make it not only the spiritual successor to the 2013 Mac Pro, but to the 2017 iMac Pro as well.
Now, the M1 Max Mac Studio starts at $2000 and the M1 Ultra at $4000, and goes up from there, way up if you ultra the GPU, memory, and storage, like $8000 up. Which is kind of a really good value on the entry level, and only harshened at the top by how low yield rates really are for two bridged dies as monolithic as M1 Ultra, even with lower end, relatively speaking, binned options, and Apple’s top shelf, but also top priced ram and SSD upgrades. Still, there’s nothing else like the Mac Studio, so if you’re not waiting on the modular Mac Pro, or holding out some last vestige of hope for an iMac Pro, RIP, and neither need these capabilities to make way more money, or just want them because you already have way more money, then have it. Personally, I ordered an Ultra, and I haven’t bought a desktop Mac in maybe a decade.
Because I kinda love it. Like all caps love it. I love having M1 Max and M1 Ultra in an enclosure you couldn’t fit Alder Lake or Ampere in if it was twice the size and liquid nitrogen cooled. And that’s including the extra two pounds of copper, rather than aluminum, heat sync in the M1 Ultra. Because damn.
And I know I’ve been focusing on M1 Ultra here, because new and shiny, but I did a whole entire video on M1 Max when it debuted in the MacBook Pro, which I’ll link up below the like button, and if you’ve just been waiting for that chip in a desktop enclosure, than that’s pretty much what you’re getting, along with a ton more thermal envelop to let it run wild in.
And I love having it this size so I don’t have to stick it on the floor and can have all that I/O immediately within reach. And I love that it’s whisper quiet unless I’m lighting up basically all the compute engines. And I don’t care that it’s an appliance without any internally accessible PCIe expansion slots, because that’s what the Apple Silicon Mac Pro will provide, what it was always meant to provide, anyway. And I love that it’s not stuck inside a display, because at this point I’ve been beyond spoiled for mini-LED and HDR by the MacBook Pro and iPad Pro anyway, and now I can get that, when and if it comes next.