Apple introduces M4 chip
"With these improvements to the CPU and GPU, M4 maintains Apple silicon’s industry-leading performance per watt. M4 can deliver the same performance as M2 using just half the power. And compared with the latest PC chip in a thin and light laptop, M4 can deliver the same performance using just a fourth of the power."
That's an incredible improvement in just a few years. I wonder how much of that is Apple engineering and how much is TSMC improving their 3nm process.
The 256gb and 512gb models have 8gb of ram. The 1tb and 2tb models have 16gb. Not a fan of tying ram to storage.
M2's Neural Engine had 15TOPS, M3's 18TOPS (+20%) vs. M4's 38TOPS (+111%).
In transistor counts, M2 had 20BTr, M3 25BTr (+25%) and M4 has 28BTr (+12%).
M2 used TSMC N5P (138MTr/mm2), M3 used TSMC N3 (197MTr/mm2, +43%) and M4 uses TSMC N3E (215MTr/mm2, +9%).[1][2]
[1] https://en.wikipedia.org/wiki/5_nm_process#%225_nm%22_proces...
[2] https://en.wikipedia.org/wiki/3_nm_process#%223_nm%22_proces...
Why are we running these high end CPUs on tablets without the ability to run pro apps like Xcode?
Until I can run Xcode on an iPad (not Swift Playgrounds), it's a pass for me. Hear me out: I don't want to bring both an iPad and Macbook on trips, but I need Xcode. Because of this, I have to pick the Macbook every time. I want an iPad, but the iPad doesn't want me.
> M4 has Apple’s fastest Neural Engine ever, capable of up to 38 trillion operations per second, which is faster than the neural processing unit of any AI PC today.
I always wonder what crazy meds Apple employees are on. Two RTX 4090s is quite common for hobbyist use, and that is 1321 TOPS each, making two over 69 times more than what Apple claims to be the fastest in the world. That performance is literally less than 1 % of a single H200.
Talk about misleading marketing...
Seeing an M series chip launch first in an iPad must be result of some mad supply chain and manufacturing related hangovers from COVID.
If the iPad had better software and could be considered a first class productivity machine then it would be less surprising but the one thing no one says about the iPads is “I wish this chip were faster”
I get frustrated seeing this go into the iPad and knowing that we can't get a shell, and run our own binaries there. Not even as a VM like [UserLAnd](https://userland.tech). I could effectively travel with one device less in my backpack but instead I have to carry two M chips, two displays, batteries, and so on...
It's great to see this tech moving forward but it's frustrating to not see it translate into a more significant impact in the ways we work, travel and develop software.
Call me crazy, but I want all that power in a 7" tablet. I like 7" tablets most because they feel less clunky to carry around and take with you. Same with 13" laptops, I'm willing to sacrifice on screen real estate for saving myself from the back pain of carrying a 15" or larger laptop.
Some of this is insanely impressive. I wonder how big the OS ROM (or whatever) is with all these models. For context, even if the entire OS is about 15GB, in order to get some of these features locally just for an LLM on its own, its about 60GB or more, for something ChatGPT esque. Which requires me to spend thousands on a GPU.
Apologies for the many thoughts, I'm quite excited by all these advancements. I always say I want AI to work offline and people tell me I'm moving the goalpost, but it is truly the only way it will become mainstream.
I love when Gruber is confidently wrong https://daringfireball.net/linked/2024/04/28/m4-ipad-pros-gu...
Only iPad Pro has M4? Once upon a time during the personal computer revolution in the 1980s, little more than a decade after man walked the moon, humans had sufficiently technologically developed that it was possible to compile and run programs on the computers we bought, whether the computer was Apple (I,II,III, Mac), PC, Commodore, Amiga, or whatever. But these old ways were lost to the mists of time. Is there any hope this ancient technology will be redeveloped for iPad Pro within the next 100 years? Specifically within Q4 of 2124, when Prime will finally offer deliveries to polar Mars colonies? I want to buy an iPad Pro M117 for my great-great-great-great-granddaughter but only if she can install a C++ 212X compiler on it.
I've got a Mac Pro paperweight because the motherboard went. It's going to the landfill. I can't even sell it for parts because I can't erase the SSD. If they didn't solder everything to the board you could actually repair it. When I replace my current Dell laptop, it will be with a repairable framework laptop.
> M4 makes the new iPad Pro an outrageously powerful device for artificial intelligence.
Isn’t there a ToS prohibition about “custom coding” in iOS? Like, the only way you can ever use that hardware directly is for developers who go through Apple Developer Program, which last time I heard was bitter lemon? Tell me if I’m wrong.
Generally, I feel that telling a company how to handle a product line as successful as the iPads doesn't make much sense (what does my opinion matter vs their success), but I beg you, please make Xcode available on iPad OS or provide an optional and separate MacOS mode similar to Dex on Samsung tablets. Being totally honest, I don't like MacOS that much in comparison to other options, but we have to face the fact that even with the M1, the iPads raw performance was far beyond the vast majority of laptops and tablets in a wide range of use cases, yet the restrictive software made that all for naught. Consider that the "average" customer is equally happy with and, due to pricing, generally steered towards the iPad Air, which are great devices that cover the vast majority of use cases essentially identical to the Pro.
Please find a way beyond local transformer models to offer a true use case that differentiates the Pro from the Air (ideally development). The second that gets announced, I'd order the 13-inch model straight away. As it stands, Apple's stance is at least saving me from spending 3,5k as I've resigned myself to accept that the best hardware in tablets simply cannot be used in any meaningful way. Xcode would be a start, MacOS a bearable compromise (unless they start to address the instability and bugs I deal with on my MBP, which would make MacOS more than just a compromise), Asahi a ridiculous, yet beautiful pipedream. Fedora on an iPad, the best of hardware and software, at least in my personal opinion.
"The M4 is so fast, it'll probably finish your Final Cut export before you accidentally switch apps and remember that that cancels the export entirely. That's the amazing power performance lead that Apple Silicon provides." #AppleEvent
Apple was really losing me with the last generation of intel macbooks but these m class processors are so good they've got me locked in all over again
Nano-Texture
I really hope this comes to all Apple products soon (iPhones, all iPads, etc).
It's some of the best anti-reflective tech I've seen that keeps color and brightness deep & bright.
The heck do I do with an M4 in an iPad? Scroll hacker news really really fast?
Apple needs to reinvest in software innovation on the iPad. I don't think my use case for it has evolved in 5 years.
Key values in the press release:
- Up to 1.5x the CPU speed of iPad Pro's previous M2 chip
- Octane gets up to 4x the speed compared to M2
- At comparable performance, M4 consumes half the power of M2
- High-performance AI engine, that claims 60x the speed of Apple's first engine (A11 Bionic)
As an engineer, I find it extremelly frustrating to read Apple’s marketing speak. It almost sounds like ChatGPT and StarTrek techno-babble. Engineers cannot stomach reading the text, and non engineers wont bother reading it anyway.
Whats wrong with plain old bullet-points and sticking to the technical data?
120 GB/s memory bandwidth. The M4 Max will probably top out at 4x that and the M4 Ultra at 2x that again. The M4 Ultra will be very close to 1TB/s of bandwidth. That would put the M4 Ultra in line with the 4090.
Rumours are that the Mac Studio and Mac Pro will skip M3 and go straight to M4 at WWDC this summer, which would be very interesting. There has also been some talk about an M4 Extreme, but we've heard rumours about the M1 Extreme and M2 Extreme without any of those showing up.
I'm extremely happy with my M1 iPad.
The only real issue is aside from the screen eventually wearing out ( it already has a bit of flex), I can't imagine a reason to upgrade. It's powerful enough to do anything you'd use an iPad for. I primarily make music on mine, I've made full songs with vocals and everything ( although without any mastering - I think this is possible in Logic on iPad).
It's really fun for quick jam sessions, but I can't imagine what else I'd do with it. IO is really bad for media creation, you have a single USB C port( this bothers me the most, the moment that port dies it becomes E Waste), no headphone jack...
Well at the end of the day the processors are bottlenecked by its OS. What real value does an iPad bring that a typical iPhone + Mac combo misses? (Other than being a digital notebook…)
Maybe I'm getting blasé about the ever improving technical capabilities, but I find the most astounding thing is that the M4 chip, an OLED screen on one side, and aluminium case on the other can fit in 5.1mm!
>> And with AI features in iPadOS like Live Captions for real-time audio captions, and Visual Look Up, which identifies objects in video and photos, the new iPad Pro allows users to accomplish amazing AI tasks quickly and on device. iPad Pro with M4 can easily isolate a subject from its background throughout a 4K video in Final Cut Pro with just a tap, and can automatically create musical notation in real time in StaffPad by simply listening to someone play the piano. And inference workloads can be done efficiently and privately...
These are really great uses of AI hardware. All of them benefit the user, where many of the other companies doing AI are somehow trying to benefit themselves. AI as a feature vs AI as a service or hook.
Why does a tablet have a camera bump!? Just take out the camera. And let me run VSCode and a terminal.
> while Apple touts the performance jump of the 10-core CPU found inside the new M4 chip, that chip variant is exclusive to the 1 TB and 2 TB iPad Pro models. Lower storage iPads get a 9-core CPU. They also have half the RAM
https://9to5mac.com/2024/05/07/new-ipad-pro-missing-specs-ca...
I understand that they have delayed the announcement of these iPads until the M4 is ready, otherwise there is nothing interesting to offer to those who have an iPad Pro M2. I don't see the convenience of having a MacBook M3 and an iPad M4. If I can't run Xcode on an iPad M4, the MacBook is the smartest option; it has a bigger screen, more memory, and if you complement it with an iPad Air, you don't miss out on anything.
if only macOS would run on iPad..
> M4 has Apple’s fastest Neural Engine ever, capable of up to 38 trillion operations per second, which is faster than the neural processing unit of any AI PC today.
How useful this is for free libraries? Can you invoke it from your Python or C++ code in a straightforward manner? Does it not rely on proprietary drivers only available for their own OS?
Today’s Xcode 15.4 RC suggests that “Donan” is the M4 core codename, and that it may support ARM’s SME instructions.
So long as it lets me play some of the less-intense 00's-10's era PC games in some sort of virtualization framework at decent framerates one day, and delivers great battery life as a backend web dev workstation-on-the-go the next, it's a good chip. The M2 Pro does.
Does anyone know how much of this giant leap performance as Apple puts it is really useful and perceived by end users of iPad. I am thinking gaming, art applications on iPad. What other major ipad use cases are out there that need this kind of performance boost.
LLaMA 3 tokens/second please, that's what we care about.
I’m far from an expert in Apple silicon, but this strikes me as having some conservative improvements. And in-depth info out there yet?
All I want is more memory bandwidth at lower latency. I've learnt that's the vast majority of felt responsiveness today. I could care less about AI and Neural Engine party tricks, stuff I might use once a day or week.
> capable of up to 38 trillion operations per second
Assuming these are BF16 ops, by comparison, an H100 from NVIDIA will do 1979 teraFLOPS BF16.
So this "Neural Engine" from Apple is 50x slower than an H100 PCIe version.
Only in the new iPads though, no word when it'll be available in Macs.
They are talking about iPad Pro as the primary example of M4 devices. But iPads don't really seem to be limited by performance. Nobody I know compiles Chrome or does 3D renders on an iPad.
So is the iPad mini abandoned due to the profit margins being too small or what? I wish they'd just make it clear so I could upgrade without worrying a mini replacement will come out right after I buy something. And I don't really understand why there are so many different iPads now (Air/Pro/Standard). It just feels like Apple is slowly becoming like Dell... offer a bunch of SKUs and barely differentiated products. I liked when Apple had fewer products but they actually had a more distinct purpose.
~38 TOPS at fp16 is amazing, if the quoted number if fp16 (ANE is fp16 according to this [1] but that honestly seems like a bad choice when people are going smaller and smaller even at the higher level datacenter cards so not sure why apple would use it instead of fp8 natively)
[1]: https://github.com/hollance/neural-engine/blob/master/docs/1...
I'm still rocking an iPad 6th generation. It's a video consumption device only. A faster CPU doesn't enable any new use cases.
The only reason is the consumer's desire to buy more.
And here I am with my MB Air M1 with no plan to upgrade whatsoever because I don't need to...
(yes, I understand this is about iPad, but I guess we'll see these M4 on the MB Air as well ?)
Interesting, it seems Apple knows they are up against the wall where there isn't really much more their devices NEED to do. The cases that Apple gives for using these devices incredible compute is very marginal. Their phones computers and iPads are fantastic and mainly limited by physics. I have basically one of everything, love it all, and do not feel constrained by the devices of the last 3-4 years. Vision and Watch still leave room for improvement, but those are small lines. Limited opportunity to innovate, hence the Vision being pushed to market without a real path forward for customer need/use. Very few people read about the latest iPad m4 and think "oh wow my current iPad cant do that..."
Curious what steps they will take, or if they shouldn't just continue returning large amounts of cash to shareholders.
I got somewhat accustomed to new outrageous specs every year, but reading near the end that by 2030 Apple plans to be 'carbon neutral across the entire manufacturing supply chain and life cycle of every product' makes me hope one day my devices are not just a SUV on the data highway.
Did they mention anything about RAM?
So why should I buy any Apple Laptop with M3 chip now (if I'm not in hurry)? lol
Given that recent Apple laptops already have solid all-day battery life, with such a big performance per watt improvement, I wonder if they'll end up reducing how much battery any laptops ship with to make them lighter.
Who would buy a MacBook Air or mini or studio today with its older chips?
I wonder if they've implemented any fixes or mitigations for GoFetch (https://gofetch.fail/).
I'm most interested in what this means for the next Vision device.
Half the power budget could well translate to a very significant improvement in heat on the device, battery size and other benefits. Could they even dispense with the puck and get the battery back onto the headset for a consumer version that runs at slightly lower resolution and doesn't have the EyeSight feature?
If they could do that for $2000 I think we'd have a totally different ball game for that device.
I would want to know if the LPDDR6 rumors are substantiated, i.e. memory bus details.
https://forums.macrumors.com/threads/lpddr6-new-beginning-of...
If M4 Max could finally break the 400GBps limit of the past few years and hit 600 GBps, it would be huge for local AI since it could directly translate into inference speedups.
> M4 has Apple’s fastest Neural Engine ever, capable of up to 38 trillion operations per second, which is faster than the neural processing unit of any AI PC today.
I didn't even realize there is other PC-level hardware with AI-specific compute. What's the AMD and Intel equivalent of Neural Engine? (not that it matters since it seems the GPU where most of the AI workload is handled anyway)
I love these advances and I really want a new iPad but I can't stand the 10"+ form factor. When will the iPad Mini get a substantial update?
Seems like we are clearly in the _“post peak Apple”_ era now. This update is just for the sake of update; iPad lineup is (even more) confusing; iPhone cash-cow continues but at slower growth rate; new product launches - ahem! Vision Pro - have seen low to negligible adoption; marginal improvements to products so consumers are holding out longer on to their devices.
I own M1, A10X and A12X iPad Pros. I have yet to see any of them ever max out their processor or get slow. I have no idea why anyone would need an M4 one. Sure, it's because Apple no longer has M1s being fabbed at TSMC. But seriously, who would upgrade.
Put MacOS on iPad Pro, then it gets interesting. The most interesting thing my ipad pros do are look at security cameras or read ODB-II settings on my vehicle. Hell, they can't even maintain an SSH connection correctly. Ridiculous.
I see Apple always show videos of people editing video on their iPad Pro. Who does that??? We use them for watching videos (kids). One is in a car as a mapping system - that's a solid use case. One I gave my Dad and he did know what to do with it - so its collecting dust. And one lives in the kitchen doing recipes.
Functionally, a 4 year old Chromebook is 3x as useful as a new iPad Pro.
imaging a device so powerful as this new ipad, yet so useless. it baffles me that we have this great hardware only the software bit is lacking
Looking forward when these new 10 core M4s end up in cheap mac desktops. I hope theres a max ram boost from the existing 24GB.
Unfortunate that they got rid of the SIM card slot, Google Fi only supports physical sims for their "data only" sim feature.
I am awaiting the day when a trillion transistors will be put on a mobile device chewing 5W of peak power.
It's going to be a radical future.
8GB RAM?? what is this, 2005? Seriously, how much would it cost Apple to start its base M4 model with 16GB RAM.
I really really hope [but doubt] Qubes OS runs on something like this with these caveats:
-using an add-in video card for dom0 gpu
-apple's on-die gpu can be passed through to a qube
-apple's unified memory does not need OS X to adjust the RAM/VRAM wall
Since I doubt these caveats are in play, I think I may be stuck with enterprise hand-me-downs that I can pass through.
almost went for M2 128gb to run some local llamas
glad I held out. M4 is going to put downward pressure across all previous gen.
edit: nvm, AMD is coming out with twice the performance of M4 in two months or less. If the M2s become super cheap I will consider it but M4 came far too late. There's just way better alternatives now and very soon.
38 TOPS in the Neural Engine comes dangerously close to the Microsoft requirement of 40 TOPS for "AI PCs".
What's the endgame with iPads though? I mainly use it for consumption, taking notes and jotting annotations on PDFs. Well, it's a significant companion for my work, but I cannot see if I've any reason to upgrade from iPad Air 5, especially given the incompatibility of the Pencil 2nd gen.
The only reason I'm buying a new iPad Pro is the screen and because the battery on my 2021 iPad Pro is slowly dying.
I could care less that the M4 chip is in the iPad Pro ... all I use it for is browsing the web, watching movies, playing chess, and posting on Hacker News (and some other social media as well).
> the latest chip delivering phenomenal performance to the all-new iPad Pro What a joke. They have M4 and they still run iOS? Why can't they run MacOS instead?
If you take it a bit deeper: if an iPad would have keyboard, mouse and MacOS → it would basically be a 10/12 inch macbook.
I dread using a Mac for any serious work, you lose a lot of the advantages of Linux (proper package and window management, native containers, built-in drivers for every hardware out there, excellent filesystems support, etc). And you get what exactly?
Give me Mac OS bootup on iPad and I am in
Don’t get the hype of the performance and being locked to iPad ecosystem
“M4 has Apple’s fastest Neural Engine ever, capable of up to 38 trillion operations per second”
Let that sink in.
Who built the M4 chip? https://www.linkedin.com/posts/jason-salt_ai-apple-semicondu...
All this powerful hardware on a laptop computer is like driving a Ferrari at 40 mph. It is begging for better use. If apple ever releases an ai robot that's going to change everything. Long ways to go, but when it arrives, it will be chatgptx100.
I hate how apple tends to make statements about their products without clear benchmarks.
Why does this feel like it was hastily added at the last minute? Developing a chip like the M4 presumably takes years, so hastily incorporating 'AI' to meet the hype and demand could inevitably lead to problems.
Cool. Too bad it’s Apple and it’s locked down to Apple products.
If Apple ever decides to pivot business to chip manufacturing that allows open access and sale. Then I would revisit.
Always need competition in this space. Especially for low power cpu/gpus.
All this power and I still can't play decent games on my macbook.
lol I just got a m3 max and this chip does more than 2x tops than my NPU does
When this arrives in MacBooks, what would that mean in practice? Assuming base M4 config (not max, not ultra - those were already powerful in earlier iterations), what kind of LLM could I run on it locally?
This is great but why even bother with the M3?
The M3 Macs were released only 7 months ago.
Looks like their NPU (aka ANE) takes up about 1/3 of the die area of the GPU.
Would be interesting to see how much they’re _actually_ utilizing the NPU versus their GPU for AI workloads.
Why does Apple hurry to push M4 before A18 Pro? Who can support the hypotheses below? 1) M3 follows M2 and A16 Pro in part, and 2) M4 follows M2 and A17 Pro.
Is there an sdk to work on gaming with unity with ipad Pro m4?
Dear Apple, please stop focusing on thinner. At this point you're selling high-end breakables. To quote Katherine the Great, Huzzah!
It's hilarious how they still push the "better for the Environment" garbage when they industrially destroy old iphones.
Anyone know if the ram multiples of the M4 are better than the M3? Example: could a base model M4 sport more than 24GB of ram?
> 3-nanometer technology
Wow. I remember being assured that we would never reach even low double-digit nanometer processes.
I always wonder how constraining it is to design these chips subject to thermal and energy limitations. I paid a lot of money for my hardware and I want it to go as fast as possible. I don't want my fans to be quiet, and I don't want my battery life to be 30 minutes longer, if it means I get more raw performance in return. But instead, Apple's engineers have unilaterally decided to handicap their own processors for no real good reason.
part of me wonders if the reason apple went with an unusual double jump in processor generation is that they are fearing or at least trying to delay comparison with other desktop class arm processors. wonder if mac lineup will get m4 at all or start with m4 pro or something. we'll see.
Anyone know more about the "NPU" (neural processing unit)? Basically a GPU specialized for AI?
Looks great. Now put it in a real computer. Such a waste to be in a jailed device that can’t run anything.
Any idea when M4 will be in a mac pro?
Why are all the comparisons with the M2? Apple did this with the M3 -> M1 as well right?
This is pretty awesome. I wonder if it has a fix for the the GoFetch security flaw?
1Tb model = €2750.
For iPad, not an MBP laptop.
Its really saying something about how the tech sector has shifted due to the recent AI wave that Apple is announcing a chipset entirely apart from a product.
This has never happened to my knowledge in this companies history? I could be wrong though, even the G3/G4s were launched as PowerMacs.
Am I wrong, or is raytracing on an iPad an _insane_ thing to announce? As far as I know, raytracing is the holy grail of computer graphics.
It's something that became viable on consumer gaming desktops just a few years ago, and now we have real-time ray tracing on a tablet.
Any hope for a new iPhone SE? My 1st gen's battery is near dead.
Am I missing it? 3x faster than an Xbox 10 from 2020 sounds "just" like Moore's law growth
Cool! Thanks Apple!
I guess this is to be expected given your market position and billions in the bank?
Thanks so much for doing it. We'll definitely buy all your
Can anyone explain where the media engine resides and runs?
> M4 makes the new iPad Pro an outrageously powerful device for artificial intelligence
Yeah, well, I'm an enthusiastic M3 user, and I'm sure the new AI capabilities are nice, but hyperbole like this is just asking for snark like "my RTX4090 would like a word".
Other than that: looking forward to when/how this chipset will be available in Macbooks!
Everyone seems as confused as I am about Apple's strategy here. I wasn't sure the M4 existed, now it can be bought in a format noone wants. How will this bring in a lot of revenue?
16GB RAM ought to be enough for anybody! (Tim Cook)
Are there enough cores to allow user switching?
Is it just me or is there not a single performance chart here? Their previous CPU announcements have all had perf-per-watt charts, and that's conspicuously missing here. If this is an improvement over previous gens, wouldn't they want to show that off?
No numbers on battery life improvements?
Bring on AI art, music, & games!
Who wrote this? "A fourth of the power", what happened to a quarter of the power?
Posts like these directly to the Apple PR dept is just an ad IMO.
any benchmarks of how it stacks up to m1, m2 and m3?
“The next-generation cores feature improved branch prediction, with wider decode and execution engines for the performance cores, and a deeper execution engine for the efficiency cores. And both types of cores also feature enhanced, next-generation ML accelerators.”
They keep making iPads more powerful while keeping them on the Fisher-Price OS and then wonder why no one is buying them for real work.
Who in their right mind will spend $1300-$1600 on this rather than a MacBook Pro?
Apple Pencil Pro...
Apple Pencil Ultra next?
Apple Pencil Ultra+
Apple Pencil Pro Ultra XDR+
Sorry to be a noob, but does anyone have a rough estimate of when this m4 chip will be in a macbook air or macbook pro?
4P cores only ???.
[dead]
yes but will it blend?
[dead]
Why even have an event at this point? There's literally nothing interesting.
No mention of battery life. They keep making stuff thin, and unupgradeable. What's the point of buying an apple device that is going to wear out in 5 years?
Are these M-class chips available to be purchased on Digi-Key and Mouser? Do they have data sheets and recommended circuitry? I’d love to play with one just to see how difficult it is to integrate compared to, say, an stm8/32 or something.
Who built the M4 chip: https://www.linkedin.com/posts/jason-salt_ai-apple-semicondu...
As an amateur EE it is so annoying that they reuse names of already existing ARM chips.
ARM Cortex-M4 or simply M4 is quite popular ARM architecture. I am using M0, M3 and M4 chips from ST on a daily basis.
Together with next-generation ML accelerators in the CPU, the high-performance GPU, and higher-bandwidth unified memory, the Neural Engine makes M4 an outrageously powerful chip for AI.
In case it is not abundantly clear by now: Apple's AI strategy is to put inference (and longer term even learning) on edge devices. This is completely coherent with their privacy-first strategy (which would be at odds with sending data up to the cloud for processing).
Processing data at the edge also makes for the best possible user experience because of the complete independence of network connectivity and hence minimal latency.
If (and that's a big if) they keep their APIs open to run any kind of AI workload on their chips it's a strategy that I personally really really welcome as I don't want the AI future to be centralised in the hands of a few powerful cloud providers.