This was discussed before and interesting but apparently the name of that instruction is misleading. Someone had chimed in and talked about how having Javascript in its name is entirely unnecessary as that exact same floating point representation is commonly used outside Javascript as well.
If you disassemble some armv8 binaries that aren't dealing with Javascript, you do see still see FJCVTZS.
There are already specialized instructions in the Apple Silicon chips. IIRC there's something tailored for the Objective-C runtime, and something useful for Javascript runtimes.
Uncontended acquire-release atomic operations are basically free on Apple Silicon, which synergizes with the Objective-C (and Swift!) runtimes, where every retain/release is an atomic increment/decrement.
> Uncontended acquire-release atomic operations are basically free on Apple Silicon
While I don't doubt you, the poster, specifically, how is this possible? To be clear, my brain is x86-wired, not ARM-wired, so I may have some things wrong. Most of the expense of atomic inc/dec is "happens before", which essentially says before the current core reads that memory address, it will be guaranteed to be updated to the latest shared value. How can this be avoided? Or is it not avoided, but just much, much faster than x86? If the shared value was updated in a different core, some not-significant CPU cycles are required to update L1 cache on current current with latest shared value.
In the future, we’ll decide HTML, CSS, and JS are too much of an inconsistent burden; so every website will bundle their own renderers into a <canvas> tag running off a WASM blob. Accessibility will be figured out later - just like it was for the early JavaScript frontends.
I am looking forward to the HTML Frameworks explosion. You thought there were too many JS options? Imagine when anyone can fork HTML.
<canvas> is already a middle finger in the direction of accessibility. You don't need wasm to put blind people at an extra disadvantage. SVG Accessibility anyone? No? What a surprise. Classical web accessibility has basically ended. We (blind people) are only using sites which are sufficiently old to be still usable.
I'm genuinely trying to do something about <canvas> element accessibility. Whether it's enough ...? Probably not. But if I can do the work to try and show that <canvas> elements can be made more accessible, then there's no excuse for developers working on far more popular JS canvas libraries from making an attempt to better my efforts.
I do strongly agree that <canvas> elements should not be used to replace HTML/CSS! My personal web hierarchy is 1. HTML/CSS/images; 2. Add (accessibility-friendly) JS if some fancy interaction is useful; 3. More complex - try SVG/CSS; 4. use <canvas> only if nothing else meets the project requirements.
Well, I am not a web dev... At least, my know-how ends when SPAs begin. All I can point you to are the WCAG, but I am sure you already know about them...
Regarding the vague criticism you mention, I'd need something more concrete to tell you if the rumors are truish...
There has been some exploration around developing a JavaScript API for accessibility. If implemented, that would allow <canvas> renderers to be accessible. I hope people will consider that blocking for shipping canvas renderers, but we'll see.
Why stop there? LLMs will free us from the shackles of having to ship actual code, instead we'll ship natural language descriptions and JIT them at runtime. It may use orders of magnitude more resources and only work half of the time but imagine the Developer Velocity™
The LLM created code will then be consumed by my AI agent which will rewrite the application to filter all of the bullshit and be fit for my minimalist preferences like a Reader Mode for CRUD apps.
In fact, with AI becoming more powerful, the <canvas> tag might soon become even more viable; because nobody will need ARIA tags or similar to tell them what’s on screen. The AI screen reader will look at the website as a whole and talk to the user. With accessibility no longer required, and with any UI being just a dumb framebuffer, we’ll finally see perfect chaos.
And blind people will be the first test subjects for the "we see everything you read" project. Sweet. A small enough group that has no way out. Besides, after the initial giveaways, imagine the revenue if you can charge for every single pageview.
The state of web deployment in 2025 is the universe punishing me for calling java applets and other java web deployment tech "heavyweight" back in the day.
Not that I intend to scale this in any way, but I'm working on an in-game UI rendered on the canvas, and I am thinking I might be able to hack something together based on this youtuber's library and excellent explainer video[0]. The thought had definitely occurred to me that if someone wanted to really roll up their sleeves and maintain a js port of the library, it would provide a translate-able UI from native C to native JS and back. At least, I can imagine a vite/webpack-like cli that reads the C implementation and spits out a js implementation.
Of course, I could also imagine one that reads the C and provides the equivalent html/css/js. And others might scoff "why not just compile the whole C app into wasm", which would certainly be plenty performant in a lot of cases. So I guess I don't know why it isn't already being done, and that usually means I don't know enough about the problems to have any clue what it would actually take to make such things.
In any case, I'm also looking forward to a quantum leap in web app UI! I'm not quite as optimistic that it's ever going to happen, I guess, but I can see a lot of benefit, if it did.
I'm thinking about this space now. Ideally, I want a new browser like platform with stricter security properties than browsers but better rendering out of the box capabilities.
Web Components was too verbose and nobody could figure it out. Flutter is just the beginning of the newest scheme by RAM manufacturers to bloat our RAM usage. We’ve stagnated at 8GB on midrange computers for too long.
Soon it'll be all 3D content anyway... the old world of a graph of documents is going away. The web breathed a sigh of relief when Apple's Vision Pro bombed.
Speaking of CRUD, would Apple’s on-chip memory have significant advantages for running Postgres vs a threadripper with a mobo full of ram?
It seems like vertical scaling has fallen out of fashion lately, and I’m curious if this might be the new high-water mark for “everything in one giant DB”.
Better get to the bottom of the mystery surrounding Apple's ECC patents on LPDDR ECC or you will have to make a leap of faith that your database on their chips won't wind up cruddy in a Bad Way. All we have now are assumptions and educated guesses about what they may be doing. It's also going to be an issue with AMD 395+ and nVidia+MediaTek GB10 (but I would assume NO ECC on those SoCs, based on their history).
It may only be a few mm to the LPDDR5 arrays inside the SoC, but there are all sorts of environmental/thermal/power and RFi considerations, especially on tiny (3-5nm) nodes! Switch on the numerical control machine on the other side of the wall from your office and hope your data doesn't change.
There are already big servers designed for huge single databases, for example the 8-socket Xeon types. Tbh I don't understand exactly why RAM is such a concern, but these machines have TBs of it.
I think a bigger business case is virtual machine hosting, say one of these is maxed out (8 Xeons with 56 cores ie 448 cores and 32tb of memory), say it's divided into a 1000 machines you can run each VM with 40% cpu utilization and 3gb of memory, considering many VM offerings have less RAM (and add a bit of overselling on top of it with regards to CPU) it could probably house over 2000 VM's.
I'm not sure how this would impact the server market in any way considering that epyc thread ripper has supported 4 TB for over 5 yrs now.
Is it the usual Apple distortion effect where fanboys just can't help themselves?
It's definitely a sizeable amount of RAM though, and definitely enough to run the majority of websites running on the web. But so would a budget Linux server costing maybe 100-200 bucks per month.
The question is about embedded DRAM, not trying to put a Mac in the data center. I am unaware of an apples to apples comparison here, but on the same Intel and AMD platform there can be a performance increase associated with embedded high speed LPDDR5 vs something on an SODIMM, which is why CAMM is being developed for that space.
I would be interested as well in what an on chip memory bank would do for an EPYC or similar system since exotic high performance systems are fun even if all I’ll ever touch at this point is commodity stuff on AWS and GCP.
Yeah, 512GB was a game changer for servers... with DDR3...
And that wasn’t even where it topped out, there were servers supporting 6TB of DDR3 in a single machine. DDR4 had at least 12TB in a single (quad-CPU) machine that I know of (not sure if there were any 96*256GB DDR4 configs). These days, if money’s no object, there exist boards supporting 24TB of DDR5. I think even some quad-CPU DDR2-era SKUs could do 1.5TB. 512GB is nothing.
(Not directly in response to you, just adding context.)
While I did make a couple of cosmetic edits within a few minutes of posting (before there were any replies), even the original was referring to the speed of the memory ("on-chip"), not its size.
You misunderstood my post, and I don't appreciate the tone of your reply.
I didn't appreciate you removing everything I responded to either, replacing it with something making my comment look entirely out of context.
While I believe you that you meant to write about the different performance profile of on chip memory, that's not what you did at the time I wrote my reply. What you actually did write was how 512 GB of RAM might revolutionize i.e. database servers. Which I addressed.
And if you hadn't written that, I wouldn't have written my comment either, because I'm not a database developer that could speculate on performance side-grades of such kind (less memory, but closer to the CPU)
This is ridiculous, I changed like 3 words. While I did originally mention 512GB, the context (“on-chip”) made it clear I was referring to the speed, not the size.
Know that scene from one episode of Aqua Teen Hunger Force where George Lowe (RIP) is a police officer and has his feet amputated, so he drags himself while pursuing a suspect?
Hmm, that hasn't been my experience. My Mediatek 4gb Chromebook is surprisingly snappy (and gets incredible battery life, better than my Macbook that cost 10x as much). Starts to slow down a bit if I go over a dozen tabs while having a video playing but otherwise, it's solid.
I can even use VS Code remote on it in a pinch, though that's pushing it...
Almost is a painful word in this case.
Imagine if it could actually run R1.
They'd make so much money. Falling
short by a few dozen GB is such a shame.
my first thought was, "what does it look like fully specced out, 512 GB RAM cannot be cheap"
fully specced out it's ~$15k
now I bet that'd be a fine $15k AI machine but if I wanted a CPU AI rig a cobbling of multi-core motherboards could get higher performance at a lower cost, and/or some array of used nvidia cards.
the good news is 3 or 4 years from now hardware specs such as this will be much cheaper, which is exciting
Not OP, and not my favorite language, but I don’t see how “Apple ships large amount of RAM in expensive workstation” is a legitimate complaint about any language. It isn’t even in the same universe of topics. Completely off-topic JS (and Rust!) drama permeating every single discussion isn’t something that happens in code review. It’s very much an expression of the HN community and its culture. And it’s really tiresome, especially when there are both better complaints and better topical venues for these languages and more.
Everyone knows that an Electron app consumes a lot of RAM. Take Slack for example. Running slack in a tab in a browser uses less RAM than running the Electron app for Slack.
The joke being that Apple realized that so many apps are built in Electron and made a decision to provide a shit ton of RAM to just to handle Electron. It seems very on point to the discussion
A joke is not deemed funny by everyone that hears it. Those that do enjoy it.
At this point, it's more satirical than haha funny. Electron is so bloated that it requires way more RAM than say native apps. To poke fun of its inefficiencies isn't going to win Last Comic Standing, but it is valid criticism even if attempted to be told in a humorous manner. Just because it's stuck in your craw doesn't mean the rest of us are in the same place as you, yet you are unwilling to accept that your view isn't the only view.
It’s not stuck in my craw. It permeates damn near every discussion no matter how remote the connection. It detracts from actual discussion of the actual topic in the process.
I actually almost totally agree with the perspective the “joke” comes from! I just don’t see it as a topic that warrants so frequently disrupting otherwise interesting discussion.
Something I’ve been surprised to find over the years working at software companies is just how many C++ writing, Linux using senior engineers there are who simply do not understand how allocation works and what HTOP is actually telling them.
I really think a sizable chunk of people in the “omg my RAM!” camp are basing it on vibes, backed up by a misread of reported usage.
This reminds a long time ago when I was trying to figure out why the heck my Intel Mac was allocating all my RAM and most of my swap to Preview or Chess.
I’m surprised how many people bring up “Erm that memory is actually not being used” as if there isn’t plenty of knock on effects from how memory pressure is actually done. For example, if I keep chrome open long enough my builds slowly use less and less threads because the build thinks there’s less memory available, so I have to periodically close chrome and reopen and restore last session.
It’s true reported memory allocation does not equal actual memory used and that’s very clever of everyone who brings it up, but it does actually cause real annoyances.
>so I have to periodically close chrome and reopen and restore last session.
I thought ublock was forced out of Chrome months ago... how are you people still using it? I switched back to Firefox a couple years ago already, even if it's occasionally painful.
Yeah I actually use Firefox almost exclusively, I only use Chrome when websites mysteriously don't work in firefox, 99% of the it's lack of firefox support.
But in my example I was thinking of a particular 2-month stretch where this kept biting me and I was using Chrome at that point. In terms of memory usage, Firefox is no better though (at one point it was, but not any more).
Now I'm afraid of saying "memory usage" lest someone pops out to comment "that's not how memory works" like whack-a-mole.
Maybe we're all idiots who look at some irrelevant numbers and declare the sky is falling. Or maybe we notice everything running really slowly because the computer is constantly paging.
Top-ish utilities should just be preconfigured to only show RSS unless you absolutely need to know what's virt. A lot of griping would diminish.
There are many specialized allocation patterns -- especially for larger system things like DBs, virtual machine / runtimes etc. -- that will mmap large regions and then only actually use part of it. Many angry fingers get pointed often without justification.
I think the griping occurs when things actually slow down, or system perceived available memory cripples resources. Maybe they point the finger to the wrong place due to misreading virtual, but I doubt people are getting angry when their system is running smoothly.
And this attitude "oh memory usage problems are a misreading of top" promotes poor memory management hygiene - and I think there's a strong argument that's all good in server applications / controlled environments but for desktop environments this attitude causes all sorts of knock on effects.
The buzz is all around AI and unified memory... but after editing 4K content on an M4 mini (versus my M1 Max Mac Studio), I've realized the few-generations-newer media processing in the M4 is a huge boost over the M1.
Coupled with the CPU just having more oomph, I ordered an M4 Max with 64 GB of RAM for my video/photo editing; I may be able to export H.265 content at 4K/high settings with greater-than-realtime performance...
I'm a little sad that the AI narratives have taken over all discussion of mid-tier workstation-ish builds now.
It feels a bit like we entered the “consumer grade workstation” era a while back when AMD started selling 16-core CPUs that will happily socket into run of the mill consumer motherboards and that continued with the higher end M-series SoCs.
It really is cool to see. It’s nice that that kind of horsepower isn’t limited to the likes proper “big iron” like it once was and can even be reasonably be packaged into a laptop that is decent at being mobile and not an ungainly portable-on-a-technicality behemoth.
The one thing that has me a bit bummed with this is that the Ultra, which I had planned to upgrade to, is only an M3 not an M4. Bit disappointing after waiting this long.
Not all that disappointing considering that most of the performance improvement in M4 seems to come from increased power consumption. In some applications, M4 performs worse per watt than M3.
Exactly. And especially for my incredibly mixed use case, which includes some light gaming in crossover, the gpu improvements in m4 are apparently non-trivial.
I’m sure whichever I end up with will be a pretty big upgrade over my basest of base model 32GB M1.
The thing I’m most curious about on the new machines, especially the Ultra, is the thermals. I only care about perf per watt if it becomes unfavorable enough that the fan spins up above idle during normal
Tasks. On my M1 the only way to get it to audibly spin up is to get the machine to near total load nd hold it there for some time.
Given that it's a desktop, most people should just get it with the default size and get an external thunderbolt NVMe disk. Only if you need >Thunderbolt 5 speeds (ie 80 Gbit/sec) do you really need the internal drive, and most NVMe is slower than this in any case.
I did this recently with a new Mac Mini that I set up. MacOS recently added the ability to locate the home directories on any volume. There's a somewhat hidden feature too that if you drag the Applications directory onto an external drive it will move selected apps there (the larger ones like Pages, etc.); combine that with the option in the App Store to keep large downloads on a separate disk.
So far it's been working quite well with the exception that VSCode does not seem to understand how to update itself if you keep it in the external Applications folder: every time it tries to update itself it just deletes itself instead. Moved it back into the /Applications folder and it's been fine.
Or if you don't want ugly ass external boxes cluttering up your desk.
I don't get why they couldn't be arsed to stuff a few m.2 slots in there. They could keep the main nand their weird soldered on BS with the firmware stuffed in a special partition if they want. Just give us more room!
I remember this but never looked into it enough to get what the point is. They still sell it with an M2 chip.
"includes six internal PCIe 4.0 slots for expansion. It does not support discrete GPUs over PCIe." uhhh, so in case people want an AS chip with most stuff soldered on but also really need certain PCIe cards that aren't GPUs?
Yeah, but it's either that or buying outdated hardware at this point because they don't refresh the Mac Pro. Idk why Apple didn't just make their own expansion add-on.
I don't know about thunderbolt, but the Apple Silicon macs I help my clients with have something really wrong and screwed up with how macOS or the firmware deals with USB 3.1+ external drives with constant disconnects despite sleeping hard-drives setting turned off etc. Searching on forums leads to similar issues others are having.
What brand and model of drive? This sounds similar to a hardware defect in some SanDisk Extreme SSDs; IIRC it was caused by firmware and/or overheating.
Yeah they really need to get that under control. It's a complete rip off at this point.
I don't mind them charging say $50 "Apple premium" for the fact it's a proprietary board and needs firmware loading onto the flash but the multiplicative pricing is bullshit price gouging and nothing more.
Exactly!! The prices are a result of extensive market research. Apple prices this things at a price they know people will buy it.
It's the beauty of having a product with no real competition in the market.
(BTW, I use Linux as my home and work OS But I'm a super geek and 20+ years full stack dev... not their target market, as I can handle the quirks and thousand papercuts of Linux)
Years ago, someone on Usenet explained that Apple upgrade prices are so high because they use components made from the powdered bones of Unicorns and I truly believe that is the truth.
I remember being a PC enthusiast in high school, spending my lunch hours pricing up Mac's, comparing them to market pc component prices, to laugh at the cost of addons. Seems like nothing has changed.
the Studio doesn't use nvme but it does put its storage on a removable card. The mac mini does as well. So you don't have to pay Apple for the storage you want. There are places which sell storage upgrades for the Mini and the M1 Studio, and they, of course, are cheaper than what Apple charges for the upgrade when you buy the machine. dosdude1 on youtube has some videos of this exact upgrade, and a bit of googling will help you find vendors. I am assuming that this M3 and M4 Studio will be the same, but that's not a guarantee.
I see iFixit[1] rates the storage swap for the M4 Mini as "moderate", I remember thinking that popping RAM in my old 2018 Intel Mini was harder than most laptop repairs I have done... I think I will probably settle for an external nVME enclosure when I get one.
They've obviously done the math on what percentage of Mac buyers will subscribe to what tier of iCloud storage, times how long people tend to keep each computer, then priced the local storage options above that: https://support.apple.com/en-us/108047
The first paragraph that talks about the OS itself is depressing:
>macOS Sequoia completes the new Mac Studio experience with a host of exciting features, including iPhone Mirroring, which allows users to wirelessly interact with their iPhone, its apps, and notifications directly from their Mac.
So that's their highlight for a pro workstation user.
Just be glad they didn't focus on movies, music and cute apps. Macs seems to be the only product line that continues to semi-dodge Apple's myopic media/services/social kiosk lens they now view all their other product lines through.
If that sounds too negative, compare their current vision for their products with Steve Jobs old vision of "a bicycle for the mind". iOS-type devices are very useful, but unleashing new potential, enabling generational software innovation, just isn't their thing.
(The Vision Pro is "just" another kiosk product for now, but it is hard to tell. The Mac support suggests they MIGHT get it. They should bifurcate:
1. A "Vision" can be the lower cost iOS type device, cool apps and movies product. Virtual Mac screen.
2. A future "Vision Pro that is a complete Mac replacement, the new high end Apple device, filled out spacial user interface for real work, etc. No development sandbox, Mx Ultra, top end resolution and field of view, raise the price, raise the price again, please. It could even do the reverse kind of support, power external screens that continued working like first class virtual screens, when you needed to share into the real world.
The Vision Pro should become a maximum powered post-Mac device. Not another Mac satellite. Its user interface possibilities go far beyond what Mac/physical screens will ever do. The new nuclear powered bicycle for the mind. But I greatly fear they want to box it in, "iPad" everything, even the Mac someday.)
I agree, except I wonder how they'll do this securely. Imagine if a VS Code plugin could spy on everything in front of me. Opens up a whole new level of security concerns.
They use a similar line on the MacBook Air page. If you're buying an (up to) $13,000 Mac, hopefully you already understand macOS and its features, I guess.
They didn't increase the memory bandwidth. You can get the same memory bandwidth, which is available on the M2 Studio. Yes, yes, of course you can get 512 gigabytes of uRAM for 10 grand.
The the question is if a llm will run with usable performance at that scale? The point is there's diminishing returns despite having enough uRAM with the same amount of memory bandwidth even with increased processing speed of the new chip m3 for AI.
> if a llm will run with usable performance at that scale?
Yes.
The reason: MoE.
They are able to run at a good speed because they don't load all of the weights into the GPU cores.
For instance, DeepSeek R1 uses 404 GB in Q4 quantization[0], containing 256 experts of which 8 are routed to[1] (very roughly 13 GB per forward pass). With a memory bandwidth of 800 GB/s[3], the Mac Studio will be able to output 800/13 = 62 tokens per second.
You seem like you know what you are talking about... mind if I ask what your thoughts on quantization are? Its unclear to me if quantization affects quality... I feel like I've heard yes and no arguments
There is no question that quantization degrades quality. The GGUF R1 uses Q4_K_M, which, on Llama-3-8B, increases the perplexity by 0.18[0]. Many plots show increasing degradation as you quantize more[1].
That said, it is possible to train a model in a quantization-aware way[2][3], which improves the quality a bit, although not higher than the raw model.
Also, a loss in quality may not be perceptible in a specific use-case. Famously LMArena.ai tested Llama 3.1 405B with bf16 and fp8, and the latter was only 2 Elo points below, well within measurement error.
But if you don't already know the question your asking is not at all something I could distill down into a sentence or to that would make sense to a lay-person. Even then I know I couldn't distill it at all sorry.
I returned an M2 Max studio with 96GB RAM, unquantized llama 70B 3.1 was dog slow, not an interactive pace. I'm interested in offline LLM but couldn't see how it was going to produce $3,000 ROI.
It would be really cool if there was awebsite "we there yet" for reasonable offline AI.
It could track different hardware configurations and reasonably standardized benchmark performance per model. I know there's benchmarks buried in github Llama repository.
There seems to be a LOT of interest in such a site in the comments here. There seem to be multiple IP issues with sharing your code repo with an online service so I feel a lot of folks are waiting for the hardware to make this possible.
We need a SWE-bench for open source LLM's and for each model to have 3Dmark like benchmarks on various hardware setups.
I get why he calls it a simulator, as it can simulate token output. It's an important aspect for evaluating use case if you need to get a sense of how much token output is relevant beyond the simple tokens per second text.
Yeah, I don’t think RAM is the bottleneck. Which is unfortunate. It feels like a missed opportunity for them. I think Apple partly became popular because it enabled creatives and developers.
The previous ranking article said the M3 ultra was the most powerful chip ever.
Mac ecosystem is starting to feel like the PC world. Just give me 3 options. Cheap, good, and expensive. Having to decide how many dedicated graphic cores for a teenagers laptop is impossible.
For chips, Ultra and Max are like their workstation chips or something, right? It seems expected that they should be a little more differentiated, they are specialist, aren’t they?
The way I think about it is if I buy a Max chip I'm getting the performance of the generation that will be released a year later now in the current form factor, and then some.
For example, I got the M1 Max when it was new. A year later the M2 came out. Spec-wise, the M1 Max was still a bit better than the M2 Pro in many regards. To me, getting a Max buys you some future proofing if you or your company can afford it (and you need that kind of performance). I use the Max with a lot of video work, and it's been fantastic.
My buying strategy has been the same since they started soldering RAM: buy the lowest spec CPU/GPU they offer with the amount of RAM I will need (which all but once has always been the maximum RAM they offer, which unfortunately usually means also buying the max CPU/GPU).
If you are in college or school getting a MacBook Air would be best and the size of your screen is going to have a higher impact (13 to 15 inch) than the Dedicated Graphics cores. Would advise not getting an MacBook Pro.
It'll likely be very workload dependent. The M4 Max will probably do a little better in single threaded tasks like browser benchmarks and the M3 Ultra will do better in things like video transcoding and 3D rendering.
I have an old style M1 Mac Mini on my desk and I could probably count on one hand the number of times I had to hit the power button, and Apple knows this, so they decided it wasn’t worth the machining cost to drill a hole in the back of the top shell and engineer a power button to the tolerances you’d expect.
Imagine, my Apple TV doesn’t even have a power button! My MacBook tells at me if I accidentally press it when doing a TouchID!
the full R1 takes >512GB and the 1.52bit takes >128GB. So enough for agent + app to realize a fully autonomous monolithic AGI humanoid head, potentially, but then it'll be compute limited...
Yeah I was thinking more about q6_0 or so. The q4_K_M is 404GB so you can still push it a bit higher than that. Obviously the 1.52 bit doesn't make sense.
I'm never going to pay 10k for that though. Hopefully cheaper hardware options are coming soon.
If you have a laptop, do you turn it off or just close the lid?
The mini is probably less power hungry than the macbooks (less components). I have some wifi 5/ac routers that consume more power at idle (nothing connected to them) than apple laptops.
> How often are you using the power button on your Mini? What is your use case?
Every single day, not by choice but because it's constantly waking up in sleep mode to do maintenance task then overheating and shutting down again. Something about macOS and Bluetooth devices not playing nice.
If you never turn off your computer, it makes sense that you never use the power button. But some people do turn their computers off, and for us, it's really useful to be able to turn them on again.
I'm still on a wired USB full sized keyboard from at least a decade ago, but didn't the newer keyboards see the return of the power button? Did I dream that?
I did some quick googling before answering, and from what I could find, people are generally saying that you can't power on the Mac Mini in other ways than by pushing its power button.
Even if you can power it on using a wired keyboard though, I'm certain that you can imagine people who prefer wireless keyboards but also turn their computer off.
I could have worded that more clearly. I wasn't disputing power button on the bottom is odd as much as thinking that Apple brought the keyboard power button back. Maybe it was the TouchID on the keyboard, but on their laptops that is also the power button, so possibly just an assumption on my part.
Oh, no I think you were clear enough, it's probably me that wasn't clear. I tried to find evidence that people were able to power on their Mac Mini in ways other than via the power button on the machine, such as a button on a keyboard. I couldn't find that, everyone just said that the physical power button on the machine is the only way.
Put simply, more people like the aesthetic of no visible power button than like the aesthetic of daily rebooting their computer.
If I were you, and I really couldn’t let go of that, I’d put the Mac in sleep and have it scripted to restart at e.g. 6AM each day. You get the best of both worlds. Feel like you have a “fresh” Mac every morning. Let it do its updates and whatnot behind the scenes.
I think another tier needs to be added to the Maslow pyramid for this particular class of complaint. I have had to reboot the M4 Mini on my desk a number of times now and it takes less than 3 seconds to lift the corner an inch and depress the switch.
It's a lab computer. You can tell people not to shut it off, but it's still always turned off when I try to use it. Could be being shut down via ITs management tools/policies for all I know.
GGML has a Metal (Apple's GPU interface layer) backend, yes, using MPS (Metal Performance Shaders), which are pre-baked shaders provided by Apple in a way similar to cuDNN. This is probably the most popular method for large-scale inference with modern bleeding-edge models.
There's also Apple CoreML, which is sort of like ONNX in that it provides a limited set of primitives but if you can compile your model into its format, it does good low-power edge inference using custom hardware (Neural Engine).
Apple also provide PyTorch with MPS, as well as a bunch of research libraries for training / development (axlearn, which is built on JAX/XLA, for example).
They also have a custom framework, Accelerate, which provides the usual linear algebra primitives using a custom matrix ISA (AMX), and on top of that, MLX, which is like fancy accelerated numpy with both Metal and AMX backends (and slower CPU backends for NEON and AVX).
Overall, there's a lot you can do with AI on Apple Silicon. Apple are clearly investing heavily in the space and the tools are pretty good.
Since Apple has always positioned itself as a tool for creatives, is it likely that the Mac Studio my be a good tool for AI video generation using open weight models like Alibaba Wan?
I found this part about PCIe expansions interesting: "For those who rely on PCIe expansion cards for their workflows, Thunderbolt 5 allows users to connect an external expansion chassis with higher bandwidth and lower latency."
(Maybe this is a feature that Apple has supported for a while, but I am unaware) Does this mean they will be officially supporting all PCIe devices like GPUs? Or do they only mean certain PCIe components like SSD expansions and network interfaces?
This has been a thing for quite a while. Thunderbolt is, in part, PCIe over serial (now USB C). There have been GPUs in external boxes, so it is undoubtedly possible, but I don't think they have many users.
in case people are unaware, this is more exciting than other random computer updates, since the Mac Studio is probably the best system in the world for running LLMs, as it can come with >> 100GB of ram and that's all accessible to the graphics/neural accelerator at high speed.
this new one comes with up to 512GB of unified RAM!
I'm not aware that Strix Halo has shipped on desktop. I thought it was only shipped on laptops and tablets? And Framework announced a desktop but did not ship it yet.
Have you used a 100GB model on a Mac studio? Tokens per second is single digit, I didn't find it usable at all, found myself going back to cloud APIs where 3000$ goes a much longer way
I'm looking forward to trying Nvidia's little set top box if it actually ships, should have higher memory bandwidth, but still Ill probably set up a system where I email a query with attachments and just let DeepSeek email me back once it's finished with reasoning at 10T/s
I disagree with the “best system in the world for running LLMs” claim. The Mac platform offers high memory bandwidth, but especially with large models you start to quickly run into the fact that the CPU/GPU themselves are very slow compared to discrete GPUs. For models up to 70b parameters, I’d much rather use any PC with a couple of 3090/4090/5090s, and for anything bigger I’d rather 1) use cloud services and pay by the hour, or 2) run a larger Epyc/Xeon system with more GPUs if my use case absolutely requires local/offline support.
The M3 Ultra with 512GB unified memory is a monster for AI development, while the new M4 MacBook Air makes AI features accessible to the mainstream. Apple's approach of building their own silicon shows massive dividends - they can optimize the entire stack from hardware to software for specific workloads like AI inference
If only it was built for workloads with value instead, I'm sick and tired of being unable to buy hardware that doesn't claim to have been made specifically for bullshit
I wonder what kind of sales volumes Apple sees with the desktop Mac variants compared to Macbooks. I know the Pro and Mini probably see decent sales number as servers, but I wonder just how many people are still buying iMacs or these Mac Studios.
For me, the main value proposition from Macs are in their laptop offerings.
Are you a lawyer? Why do you only ask questions to which you already know the answer?
Snark aside in case you're seriously asking, it's a PR thing. Generation to generation might not show much differences in direct comparisons that makes the crowd ooh and ahh. The M1 chip was the first Apple Silicone chip, so going back to their first one for basis of comparison provides for more oohs and ahhs. The charts look pretty this way too.
For people who upgrade annually, comparisons don’t matter.
For people who are making a forced purchase, comparisons don’t matter.
For people who are content with what they’ve got, comparing with the oldest-popular market segment offers a clear statement of improvement and helps long-term users calibrate when to make their next purchase.
Has it? These names are basically (architecture version, amount of compute) tuples. Much better than what AMD, Intel, Nvidia, etc. are currently doing.
This question is trivially answered by visiting Apple's website, which is---to my point---not generally true for their competition. If you have some further point to make, I recommend stating it more clearly so we can avoid wasting time here.
Why is the "Ultra" more powerful than the "Max"? I would expect "Max" to mean "maximum" in this context but it seems to mean "directly subordinate to that which is the maximum". This is pretty obviously a point of confusion. Just because other CPU manufacturers do goofy stuff with naming doesn't mean that Apple is exempt from criticism for doing something so obviously bereft of common sense.
Well then, it's a good thing I didn't say anywhere that they should be exempt from criticism. Incidentally, it's not something I believe. But I do think they are still doing a much better job at naming than other CPU manufacturers, despite the obnoxious Pro/Max/Ultra stuff.
I suppose I commented here because I think people are letting their subjective distaste for those terms sway their opinion of a superior naming scheme.
Why would you expect that? Also, the M4 is faster than the M3, but that doesn't mean Apple couldn't or didn't want to fuse two of them together for an M4 Ultra.
Is an Intel 10700K faster than a 12400F? The generations are different but the chips have vastly different capabilities and features.
M4 is the generation. The modifier modifies the generation. M4 Pro is an M4 with some extra pizzaz. M4 Max is an M4 with lots of extra pizzaz.
> This question is trivially answered by visiting Apple's website, which is---to my point---not generally true for their competition.
wtf are you talking about? Intel, Nvidia, and AMD all absolutely have complete specs for their products readily available on their respective websites. Much, much more complete ones than Apple does as well.
I guarantee you that a quick scan of any of those companies' websites will not equip one with a useful general understanding of the naming scheme they use for their products, in the sense that one can see a product name and immediately know where it falls in their lineup and what workloads it's meant to handle. That is what "the fuck" I am talking about.
Intel desktop options are quite simple. My only complaint is on laptops, where the i3, i5, i7 thing conflicts with U vs H, and almost seems intentionally misleading. Like, why does i7-U even exist?
But the nice thing is you search the model name, and Intel gives you all the specs upfront.
I like how the table of suffixes hasn't been updated to add "V" but the section on Core Ultra uses 288V as an example. The document's too big to stay in sync with itself.
The bit I found confusing is that I don't immediately understand what's more performant in various scenarios, M4 Max or M3 Ultra. The former has higher architecture version but lower amount of compute, the latter is previous arch but more compute.
Different tiers within the same generation are much easier to compare than across gens. Apple could benchmark them all and name them accordingly, but even that's misleading because it can be workload-dependent.
Fair, but it's not like other CPU/GPU manufacturer naming schemes give you that either. At least Apple's scheme clearly tells you e.g. "previous arch but more compute".
It’s definitely a little odd to have M3 Ultra > M4 Max, but I feel like anyone complaining about this must have never bought any other manufacturers’ wares in their lives. Obtuse complication is kind of the norm in this industry.
Oh, but things were far worse back then, in terms of knowing what you were getting. For instance, let's say you bought a 13" MacBook Pro in 2016. Do you have a dual core or quad core processor? Depends on whether you have a touchbar!
(For reasons best known to themselves, Apple made two completely different 13" MBPs that year, both new, with the loathed butterfly keyboard, weighing a different amount, with different processors, and the same name.)
Literally not even the same thing you're comparing. There were more processor and graphics options. Want an i5, i7, or i9 and what about a RX 580, 5300, 5500, 5700, or 5700XT?
Keep these things the hell away from the people who develop Chrome and desktop JS apps.