> Additionally, support for up to 128GB of memory unlocks workflows previously not possible on a laptop, such as AI developers working with even larger transformer models with billions of parameters.
The presentation explicitly highlighted this very randomly, without showing how AI development actually works.
I know Apple has put more effort into Apple Silicon support for PyTorch but is it there yet?
Consumer nVidia cards are constrained by their RAM which is inconvenient considering the sizes of popular LLMs. An M2 Ultra basically gives you 100+ GB VRAM which is nice.
Although pytorch is in python, some ML features in SoTA libraries are coded against CUDA and complains if you don't have a CUDA device.
A lot of things work out of the box with pytorch on mps, but not everything, and it can be frustrating to figure what works and what doesn't while you have a bunch of other moving parts shifting under your feet due to how fast the whole thing is moving.
That said, llama.cpp (aka "ggml") on M1/M2 Mac is very robust. If you just want to run inference (as opposed to training) of models and you're comfortable with a Mac, there's really no reason to go out and by an nVidia card.
Especially since (and this just occurred to me), in a roundabout way the original may have been the first Mac game to ever use raytracing (in its pre-rendering)
I'll admit "evil" is a stretch but I was thinking in terms of the amount of e-waste this causes due to insufficient RAM hurting the longevity of a $~2k+ computer
Hyperbole like this is necessary for people to justify the extreme negative emotions they feel about a computer they’ll never buy having specifications that they disapprove of.
Is it? Its pretty evil to overcharge and prey on a customer base who doesn't know any better, but would benefit from a more usable starting point, oh say 16...
Tims a finance guy, they are kinda evil by default.
Yeah, Apple customers tend to also just "trust Apple" and get whatever the recommended product is for their price point. Half the reason I find this troubling is ewaste, the other half is I'm sick of having to try to explain RAM to my MIL and ask how many tabs they keep open when asked which computer she should buy. If they had a better baseline memory (16gb) or a simple inexpensive upgrade that's the second option in the lineup instead of the higher storage cost, it'd be a simple, one-sentence "get that one".
Even having to upgrade the RAM is a hassle when price comparing etc. They know this and it's - by design - misleading to the type of affluent less-tech-friendly users who prefer Apple products.
No one should have to care how many tabs they have open, and since browsers kill/suspend background tabs these days, they probably don't have to care about it either.
America is truly a blessed nation. Their richest evil people simply make the most powerful consumer appliances available for a reasonable price. Their good people must be truly saints.
Who? As far as I can tell every American is a psychopathic capitalist murder ready to drive their SUV over the next sand dune/forest/oil field/grandmas house you name it. Bought and sold in 1775 boy.
It's a matter of perspective of course. I think if you claim that this makes Apple evil then you would also need to argue that most of capitalism is evil. Aren't the vast majority corporations just trying to make more money? I would disagree with that.
Im just some punk trying to have a good time with some spicy comments on HN. It's ok to both hate and exist within capitalism. It requires a certain level of pain (masochism) to survive that existence, or dare to enjoy it, while also raging against it.
8gb is pretty bad. I have an M2 Pro with 16gb, & that limit is regularly pushed, & that without regularly pushing things hard, resource-wise (then again, I've seen plenty of people on both PCs & Macs shut down whatever applications they weren't actively using when they switched between them. If testing shows that to be really common, I could see 8gb being plenty for the majority of users. Crazy user patterns, to be sure, but to each their own). If I get an M3, it will have 36 (or whatever Minis will have available) just for the headroom.
Literally the only reason I can’t use my work MacBook for any work that isn’t just testing Safari support. I was even ready to tolerate the terrible ram quantity given it’s so friggin’ insanely fast at javascript.
Well the average VAT in Europe is about twice as high if not more than the average sales tax in the US.
So I’d rather have to do some basic math but save 10% (of course different tax rates only partially explain the price difference, it still is significantly more expensive in Europe without VAT)
Why is it ridiculous to allow for tax-excluded display of pricing? You couldn't run a TV or newspaper ad with tax-included pricing in the US since there are so many different taxing jurisdictions here (more than 13,000, as of early 2023).
Perhaps what's ridiculous is that we have so many different taxing jurisdictions.
I think this is brilliant: let the people know how much of the price actually goes to the company, and how much to the government. Helps people make an informed decision on who's to blame for high prices of everything.
Here's some example US->EU Markups for other laptops in the same (EU) price range, sold direct by manufacturers to Ireland (list price to list price, so tax excl in US, tax incl in EU):
Yup, it seems that the "Apple tax" has gone up quite a bit since Apple shifted toward soldered RAM and storage for their products. You get excellent Apple Silicon chips with their performance and energy efficiency, as well as the most polished desktop OS (macOS), but it comes at a steep price, especially at non-base configurations. There seems to be plenty of deals for base-config Macs at places like Amazon and Costco, but if you want 24GB or 32GB of RAM, it seems that no deals can be found; you must pay full price for a custom configuration from Apple.
I was a long-time Mac user. I remember when my 2006 MacBook cost less than comparable PC laptops. My last Mac purchase was a refurbished 2013 Mac Pro I purchased from Apple in 2017; since 2021 I've switched to PCs. While I love the configuration choices and expandability of PCs, the operating system situation leaves much to be desired, in my opinion. The shenanigans Microsoft continues to pull with Windows as well as the Sisyphean development cycle of the Linux desktop ecosystem (systemd, Wayland, etc.) makes me peek at Apple whenever there's an announcement, but the steep prices for a Mac configuration that fits my needs are too much for me, and so I remain a PC user.
Definitely the main thing preventing me from moving on from Mac is how bad Windows and Linux is.
Around 6 months ago I actually bought a beefy desktop PC to play around with, dual booting both Windows and Linux and they were just terrible. The default keyboard shortcuts on Windows are a joke (not to mention, the lack of an additional modifier key which is present with mac, which makes shortcuts more difficult than it needs to be).
Linux on the other hand, was so limited. People then argue that Linux is customizable, but I spent forever trying to customise it and half the time, the modifications literally don't work at all. I assume because there are so many distros and the eco-system is so fragmented that something is almost guaranteed to break.
What is superior about Mac shortcuts? They seem to be just as inconvenient and requiring a rebind on both systems (though given how relatively easy it is for the basics strange it's a barrier to switching)
Having an additional dedicated modifier key is night and day in terms of creating shortcuts.
I don't see how they're equivalent. F2 is vastly inferior to pressing the enter key to renaming a file. Alt + F4 to close a window as opposed to cmd + w.
Can you name shortcuts that are more convenient on Windows than on Mac?
Fair enough. I guess ultimately it comes down to what you're used to and what you prefer.
I think more than anything, it's difficult to discredit something that you don't actively use and aren't completely familiar with. Like I'm just saying Windows is shit because it's not Mac, but that's disingenuous because I don't actively use Windows as my main OS. But hey, if you like Windows, use Windows. If you like Mac, use Mac.
I am sure if look closely at Windows laptops you can find plenty of competition. Of course the screen and the speakers are not as good and battery life is shorter, but if you can look over that (and I find most people outside HN don't really care) you can go pretty far with a premium Windows laptop (XPS/ThinkPad/HP EliteBook).
What's even worse are the upgrade prices: $200 to go to 16GB and $400 to go to 24GB. Anything more than that requires upgrading to the M3 Pro model, which has a base price of $1999 and a base 18GB of RAM. Upgrading to 36GB costs an additional $400.
I waited for today's announcement because I'm in the market for a laptop with 32GB of RAM. Today's announcement looked really tempting! However, $2399 + AppleCare + sales tax for the 36GB M3 Pro MacBook Pro is too much for me, and so it looks like it's either a refurbished M2, a Framework laptop, or a ThinkPad for me.
Given how much money Apple has, this is simple profiteering. Something governments should look at and come at Apple hard, given this behaviour is inflationary.
I agree and I think it's going to happen eventually. At least something is going to happen. It already looks like customers have been a bit wiser and stopped buying the most egregiously priced models. They are running their computer division to the ground with their outrageous financial mastermind strategy. They put up a good marketing front for their proprietary technology but there is no serious advantage to their stuff anymore. Most prosumer performance focused customers are switching or are thinking about it.
You just need patience and enjoy the fall. Just a few years and soon enough nobody is going to care developing stuff for their platform, it will be mostly empty shells, but at least they are pretty, I guess.
Patience my friend, karma will do its thing eventually...
I purchased a fully maxed out M2 Max MacBook Pro when it was announced like 6 months ago. I knew it was going to be outdated eventually, but within 6 months, wow. Did not expect that.
i mean, if the machine does the job you purchased it for, is it really outdated? Unless the job it is doing is to be the best and latest, as a status symbol...
Yeah, I still have an M1 Mini and it does everything I ask of it pretty handily. The most intense work being rendering edited photos out of Lightroom while watching videos and web browsing around the same time.
Oh I'm super happy with it - it has handled everything I've thrown at it and then some and I'm not planning on upgrading anytime soon. But I feel like with the M2 Pro's having such a short shelf life, what that means for long term support.
I have the same M1 Max as a personal machine and just got a maxed out M2 Max for work. The M2 Max does feel faster through daily use but that could also be related to the new install?
I don't understand why they don't update the entire lineup whenever a new chip comes out, especially if really only swapping the chip in the macbook pro and imac here.
On a separate note does anyone have any insight on how unified memory compares to vram in the context of machine learning performance? Considering an H100 with 80GB costs like $30k a maxed out macbook pro with 128gb unified memory for $5k is interesting. Is it remotely comparable or compelling for large models considering realistically most prosumers are capped at the 24gb nvidia cards for anything resembling a reasonable budget.
They just updated their macbook air and mac studios and mac pros like 6 months ago? If they updated right now, the previous owners would feel utterly screwed over.
And more importantly, these improvements are mostly thanks to TSMC's progress. You could predict the M3 already late 2022 when TSMC announced that they were starting mass production of their 3nm process. We knew that it would be a huge improvement based on the specs we had back then and roughly how long it takes to ramp up.
Only 3, unless you're counting the developer boxes? M1 machines were released in November 2020. However, M1 Pro and M1 Max were released only 2 years ago. Things are definitely picking up now...
Not even. If you ordered a non-baseline M2 Ultra Studio at launch, you likely didn't receive it til the first week of August. So we're a week shy of three months. So I'm glad for not being screwed, I think. Although who is to say they're not just being sat on (M3 Max chips)...
Right? I’ve just setup my office and for now I’m using my MBP hooked up to a couple of Studio Displays - which I adore for software development. I was planning on picking up a Mac Mini or Mac Studio so that I could still have the MBP spare.
Maybe not. It’s a bit of a pain having to unplug and replug everything in, not to mention having windows and such shuffle around each time.
Speculating, but I'd wager it's to clear out existing inventory by phasing out the old chips on other SKUs over time. E.g., you start with the expensive Macbook Pro models when you have fewer M3 chips in production, since those will sell slowly, and then you clear out your remaining M2 inventory in the Air lines while waiting for a "refresh" when M3 production is more ramped up.
I have no idea what Apple's sales are for the macbooks or if Airs sell more than Pros, etc., but they're definitely making profit this way to be this consistent in their approach.
Depends on if you're talking about training performance or inference performance. It's probably a decent deal for inference on such large models, but I doubt it's anywhere near competitive for training. In the 'consumer' space there's also the A6000 with 48GB for ~$4k.
The M2 Ultra tops out at 800GB/s unified memory bandwidth. A 4090, on the other hand, has 1,008 GB/s. On the PC side, dual-channel DDR4-6400 offers a bandwidth of 102 GB/s.
They're not really directly comparable due to the architectural differences, right? For pure rendering workflows I'm sure the 4090 is faster, but for anything that hits the CPU ...
Yes, I should've been more specific. Mobile GPUs often employed in ARM architectures often use a technique called "Tile-Based Deferred Rendering" (TBDR). This involves not just splitting the frame into tiles but also delaying the actual shading operations. This two-step approach minimizes overdraw but can create bottlenecks in scenes with intricate geometry or advanced lighting, shadows, and reflections.
How is the deferred rendering in TBDR different from the deferred shading and deferred lighting that have almost completely supplanted forward rendering in commercial games and engines?
1. Engineering bandwidth (it's work to upgrade a lineup! They probably need at least a year)
2. Manufacturing bandwidth - they probably spent the first half of the year manufacturing the 3nm iphone processor while researching the mbp processor. I expect it's difficult to ramp manufacturing on many chips simultaneously
3. Sure, consider the demand side. If my parents are going to buy a new laptop and a new phone this year, I think they would be more likely to do that 6 months apart. Similarly, I expect keeping a cadence of announce products a,b,c in this quarter and d,e,f in that quarter helps keep apple in the news (for good/exciting things)
Wrt machine learning: I can't wait for the results to come out for this once we can get our hands on it
> 2. Manufacturing bandwidth - they probably spent the first half of the year manufacturing the 3nm iphone processor while researching the mbp processor. I expect it's difficult to ramp manufacturing on many chips simultaneously
Sort of. It's that the whole concept of a "ramp" implies that manufacturing things (not just chips!) starts expensive and risky and get cheaper and reliable over time as people work out the kinks in the various processes.
The M2 is mature. TSMC can make them reliably and in high quantity at low marginal price. Why would you *not* sell a product like that if there's a market for it?
Apple introduced the 15 inch MacBook Air on June 6th.
If they upgraded it to the M3 right now that machine would have only lasted 4 months and 4 days.
Sure they could do it, but how many sales are they really losing because they haven’t? People buying the Air are not exactly looking for the absolute best in technology. They’re probably much more likely to be price sensitive, and putting a brand new M3 in would likely make that worse.
So if you consider that you have to make new boxes and manuals and new hardware and test it and change the lines and everything else… what are the chances you would come out positive after such a short amount of time?
It would make some more sense for the 13 inch air which is about 18 months old, but again if you don’t have to maybe you keep selling the M2 for another six months.
Besides wanting to clear out existing M1/M2 capacity, I don't there's enough M3 stock right now.
I forget how many fabs TSMC has that have 3-nanometer capability, but it's gotta be low. Apple's supposedly got _all_ of TSMC's capacity[0], but I doubt it's enough.
Regarding updating the lineup, I would guess that it is completely intentional to keep the carrot just out of reach across the lineup. Each new chip release rotates the product categories in just the right way to keep the consumers consuming.
It makes a world of sense if you think of it from a logistics point of view, especially considering Apple sells insane volumes of everything they sell.
It costs less to manufacture the current generation ship and the last generation chip concurrently, it allows them to push the newest generation chip out the door faster without running out of supply or having to open up excessive production lines.
>H100 vs apple silicon
while it's only one spec, H100 has over 2TB/s of memory bandwidth and Apple silicon caps out at 400GB/s. H100 in general kicks the pants out of Apple silicon performance and has far more support. These parts aren't really in the same league but Apple Silicon has become a popular hobbyist choice for LLM inferencing.
Also see: The Jetson AGX Orin 64gb, Nvidia RTX A6000 48gb, AMD Radeon Pro W7900 48gb, two nvlinked 24gb 3090's.
My gut tells me the Macbook Air sales have been cannibalizing Macbook pro sales.
For the M1 and M2 generations, the airs released in the fall and the pros a few months later in the spring. The airs have gotten so good that many people that would have gotten the more expensive (and more lucrative) pros got airs instead.
It doesn't have to be "heavy": just running a browser in a different OS may need a minimal amount of RAM (even if the 1-2 assigned vCPU/cores are mostly idle).
Bring in people with expertise, buy outside expertise where you can, focus on the product you actually want instead of the product that will appeal to every possible use case for every possible customer that the chip marketing team can imagine, don't sweat where you can put the margin line between chip and system integrator, because they are both you.
I suspect any $2,600,000,000,000 company could pull it off, if they moved first.
I don't think we're comparing apples to apples here (excuse the pun). These chips have something like 10% of the instructions that a typical x86 chip does. Once the big CPU players start producing the same kind of chips, I greatly expect Apple's power to performance advantage to drop significantly, if not be overtaken by the likes of AMD, etc.
> I greatly expect Apple's power to performance advantage to drop significantly, if not be overtaken by the likes of AMD, etc
Nope. Apple and AMD both use TSMC for manufacturing. It's all made by the same factory. AMD does not have the advantage there. Apple buys the most capacity on the most advanced process nodes since they place much bigger orders (Apple also has 10x more cash than AMD).
The iPhone chip was seriously competitive with laptops 2-3 years before the M1, it's just that due to the software environment few people noticed what was going on.
The real question is why Apple left it so long. They clearly wanted the first generation to be a clear success, but they could have probably pulled this off faster had they wanted to.
What's with the 36 GB option? The other memory configs (16 GB, 64 GB) are still clean powers of two. Size suggests that they're using ECC-capable memory but using the extra width intended to support ECC for data... but why would this only apply to a single size? Part availability?
EDIT: Digging this a bit, it's not (one or more) 72-bit wide busses with 2^32 words as I'd expect as a gray-beard, it's (probably) six 32-bit wide busses with 6 GiB per bus; and this use of 1.5 * 2^N deep memories has become relatively common with the use of IC stacking, with 12 power-of-two sized ICs stacked in a single package (instead of the more "comfortable" 8 high or 16 high stacks of the same ICs giving powers of two).
Doesn't look like it. Memory ICs are tested before packaging, and then are packaged in stacks. While I'm sure there's some loss in packaging, probably not enough to justify this level of binning; and most of the failures that would happen at this stage would potentially interfere with signal integrity for the whole stack.
Instead, it looks like after the 8-high stacks that I knew about, manufacturers went to 12-high stacks. There are 16-high stacks available too, but looks like there's a lot more backside thinning needed to keep the same Z height so cost goes up disproportionately and 12-highs are a reasonably sweet spot for capacity per dollar.
There are non power of two sized DRAM dies in production these days; it's not purely a matter of stacking (and stacking doesn't make it any easier to have something in between one die per channel and two dies per channel). You can get 24GB single-rank and 48GB dual-rank DDR5 UDIMMs compared to a year ago when the options were 16GB and 32GB; no stacking at all, just going from 16Gbit to 24Gbit per die. But LPDDR has been doing this for longer due to demand in the smartphone market.
Why do I need to select the most expensive M3 Max version (the one with 16 cores) to get 48 or 64 gigs of RAM. The M3 Max with 14 cores only allow 36 or 96 gigs. Just let me choose 14 cores and 64 gigs.
You are in their target market of who they can really squeeze since you want something a little bit special. Their memory pricing has always been absurd. Even on an older MacBook it was often +$200 for something silly like a 16GB SODIMM that costs $50 at the store.
I still think their build quality , screen quality and OS are pretty much unmatched.
It’s worse. The configurator is labeled with a relative price upgrade but an absolute RAM size. The actual deal is +$200 for +8 GB RAM which is outrageous.
Sure, but they aren’t charging you anywhere close to cost. It’s painful for a company that size to add an option like this. Plus they know they can add a chunk on because if you’re in the market for that option, they know how much you want it.
The same discussion about cost/value comes up every single time.
Often, people will search for a random part that fulfills the same function, sort by cheapest, and then let their indignation run wild, completely ignoring the difference in form factor, other properties, or even quality.
That said, it’s no secret that Apple adds a healthy margin and an “inconvenience” tax.
In Apple’s ideal world, all people would purchase a handful of mass-produced configurations. This saves them in manufacturing costs, assembly costs, and logistical costs.
Apple also spent an ungodly amount on engineering to “make more with less”.
In the long run, this saves them money on lower-capacity components, especially at the quality and with the ancillary properties they're purchased.
This is why spec for spec their iPhones and Macs look underpowered compared to competitors while performing the same if not better.
So, from their perspective, it’s “fair” to upcharge the “spec peepers” and professionals who really need it. The latter is generally less price-sensitive.
As you've stated people will make comparisons between the cheapest bottom rung slot-in ram against what's on offer from Apple's UMA. It's naive.
But even when one could just pop in your own RAM, many would still buy Apple's upgrade, which sounds insane on the surface. But there are actually pretty sound reasons for it:
1. They're buying the Mac hardware because it works. That's one of the core motivations for spending the extra to begin with.*
2. Adding ram is the gateway to unexpected crashes, such as errors that only pop up during heavy use, higher temperatures and the like. It doesn't have to be cheap either.
3. But if one is going for those cheaper components, one gets what they pay for: "mislabeled"(online fraud schemes), counterfeit, bottom-rung binned components sold as legit are par for the course.
--
* A lot of brands say their products work, when they simply don't. Here's some of my own examples:
I'm now onto my 3rd brand of mesh wifi routers - why? Because what's out there is garbage, even when you pay a lot for it. It's clear that there is insufficient QA on many top brands, and they simply won't acknowledge that the product they've sold you doesn't perform.
I purchased one of the pricy 5k LG screens that were a total lemon, at this level of expense you don't expect school-boy errors in hardware. LG handled it poorly, both in their return and exchange options, as well as customer support and patches.
I 100% agree with your comment down to your anecdotes on mesh routers and TVs.
Specifically, when it comes to routers, it seems that manufacturers only ensure that the basic routing works adequately and that all the bells and whistles exist solely for marketing reasons.
This is so bad that their CS is trained to have you disable said marketed bells and whistles during troubleshooting, only to conclude that everything is fine as long as the basic routing functionality works and that anything that doest work beyond that is a “you” problem.
But the bandwidth goes like, 100 GB/s, 150 GB/s, 300 GB/s, 400 GB/s. Given either CPU or GPU alone can't saturate bandwidth (assuming M3 the same as M1 and M2), perhaps the memory controller or the chip layout is to blame.
Right, the two M3 Max SKUs being discussed above are the ones with 300GB/s and 400GB/s memory bandwidth.
That sounds like 3 memory controllers vs 4 memory controllers, likely due to chip binning. TSMC's 3nm processes supposedly still has relatively low yields.
I'm running the lowest spec m1 with 8 gigs of ram for my daily driver.
I frequently have tens of chrome browser tabs open across multiple profiles, multiple chrome debuggers, at least two or three VS code instances, multiple high memory nodejs processes, video conferencing, screen sharing and building code all at the same time - and the thing runs smoothly without a hiccup. And does it on battery! For a full work day, at least.
Weighs less than a pixie's fart.
I can't even imagine the capabilities the new m3 max has with that kind of memory and power available.
It's going to pinch, but I guess I'll be finding out in a few weeks.
What's your swap used and memory pressure in Activity Monitor? Regarding Chromium browsers, I have noticed that they tend to unload tabs more aggressively these days.
I have the same one and it’s neigh unusable. The memory is constantly maxed out and it freezes frequently while swapping. I can’t use it. It’s in a drawer somewhere.
Just get the 96G option? It is only around 100$ more for 1.5x the memory. And 1.5x the memory is worth it when you are interested in high memory applications anyway.
What is your usecase, LLMs? Other AI models? Multiple large VMs? It’s always worth it to max the memory. In this case going from 64 to 96.
It also appears they've dropped the memory bandwidth from 200GB/s on the M1/M2 Pro to 150MB/s on the M3 Pro, and you have to upgrade to the tippy top M3 Max chip to get the full 400MB/s bandwidth experienced on the M1/M2 Max chips.
I remember getting one of those shiny aluminum MacBooks in 2009. Then MacBooks turned plastic and the aluminum ones became MacBook Pro's. Is this a thing Apple does? I'm not an Apple customer for many years now.
What you described as a transition to plastic for the MacBook was actually the final iteration of a fairly long line of plastic MacBooks, most of which were a pretty similar design to the pre-Intel iBook.
"MacBook" (not Pro) was plastic from the start, and introduced the same year as the MacBook Pro (always aluminum): 2006. The plastic MacBook was discontinued after the 2010 model. There was a one-off aluminum 13" MacBook in 2008, but it coexisted with the plastic MacBook and was not replaced by a plastic model—instead, the 13" model was promoted up to the MacBook Pro product line, which previously had been just 15" and 17".
After several years of not having a plain MacBook, Apple did the 12" aluminum MacBook from 2015 to 2017. It was basically what they wished the MacBook Air could be, but was too expensive and too thermally constrained to truly replace the MacBook Air of the time. It was killed when they finally upgraded the MacBook Air to have a Retina display.
I prefer that tbh. If I'm frequently saturating 16GB of memory in the previous gen and it is time for an upgrade, I probably don't need literally double the memory, but 24GB would be nice.
Pretty underwhelming a bit like the A16 release. They basically upped the frequency in A16 and that's how they got their single core performance improvement, not much of an IPC increase to speak of. Looks like the it will be the same story here given the performance numbers Apple gave. A16 also came with a power consumption increase so the 3nm process from TSMC is disappointing as well.
Isn't this typical tick-tock move? Don't change too many things at once - they are moving to the new process so probably not too many changes in the architecture, only increase in frequency for same power consumption allowed by the better process.
It’s kind of strange their benchmark for “Machine Learning Programmers” is “Simulation of Dynamical Systems in MATLAB.” Seems like they could have capitalized better by using a generative ai inference benchmark.
Possibly this, but also consider Apple is playing major catch up with generative AI. I could see avoiding mention just to keep it a bit further from the mind of reviewers.
I agree the machines have been good to run various ML models that aren’t running on other PCs in their category. (Though, shrinking the resource requirements seems like a never ending exercise.)
Apple likes to wait until it can announce something with a most around it. In this case, more compute isn’t enough, I don’t think.
With a MBP, sure. I don’t want a machine that large however. I have both a 14” Pro and a 15” Air that feels significantly smaller than its “smaller” counterpart…
Seemingly no. So far it looks like (someone please correct me if I’m wrong):
- M3, 1 external display
- M3 Pro, 2 external displays
- M3 Max, 4 external displays
This has to be a product differentiation decision at this point, right? Is there any serious technical limitation against more displays in the base chip, that Apple has not been able to solve in three generations?
It’s frustrating. I wish they would just charge an extra $X for the privilege of using multiple external monitors. I like the form factor of the MBA as is… if I need to compromise on that, I might as well choose a different machine.
The person you replied to wants the nicer form factor of the MBA with the ability to use multiple external displays.
That combo isn't currently offered, and judging from today's M3 specs the next MBA won't likely offer it either.
I assume the iMac has the same limitation, which is a bit of a downer for some uses. It would be nice to drive a second monitor and a wall-mounted TV or projector from an iMac in sleek a home setup.
My bad, I missed that specification. I would agree with them, it would be nice if you could pay more for a MacBook Air with multiple external display support.
Same, the M2 MBA is all the power I need and I vastly prefer the smaller size, but the monitor issue is killer.
I finally bit the bullet and bought a DisplayLink dock, which gets the job done, but just seems like a silly compromise when the Intel MBA could support 2 external monitors natively.
I hear the DisplayLink solution is very half-baked and has many issues. I'm going to skip mac laptops for as long as they don't ship a Macbook Air with dual external display support -- I am doing fine with Windows laptops, and they really have improved in performance and efficiency over the last few years.
Exactly. I imagine most of the benefit of the new 3nm process is related to power draw. I'll bet the memory is the same stuff as the prior generation though, given those specs haven't seemed to change at all.
Its on the release linked. 15% faster than m2 for P-cores and 30% for e-cores, which is misleading, because it has 2 more of them so they're not necessarily that much faster.
For me, still on m1 is good news, one more generation I can skip without feeling too much FOMO. Graphics improvement are irrelevant since im not gaming on them and cpu difference is fine for being just <40% overall.
The new M3 base model, along with only supporting one external display also loses a Thunderbolt port. A new shell / chassis just to remove a port. Courage!
>Rendering speeds are now up to 2.5x faster than on the M1 family of chips.1 The CPU performance cores and efficiency cores are 30 percent and 50 percent faster than those in M1, respectively, and the Neural Engine is 60 percent faster than the Neural Engine in the M1 family of chips.
They didn't say, but since their A17 is 3nm this is almost guaranteed to be 3nm. The 3nm process is the big improvement, the M1 and M2 were TSMC 5nm and 4nm which was basically the same thing.
The performance improvements here is roughly what you would expect if you knew the specs of TSMC 3nm chips.
Significantly faster absolute single core CPU performance than a 12900K is pretty good and not disappointing at all. I find the improvements in A17 Pro substantial.
5nm vs 3nm and three years of progress. Intel's e-cores have been seeing more generation over generation uplift compared to their p-cores so nothing about this strikes me as anything unexpected.
When the m1 came out details about the m3 were well known and that this was going to be a pretty big generation for performance improvements. I think the surprising thing was that m2 got pretty close to m3 in situations which were not as thermally constrained.
> Apple unveils M3, M3 Pro, and M3 Max, the most advanced chips for a PC
Nice! When can I order these most advanced PC chips for my next build? Do they have any kind of upgrade path for existing AMD/Intel users? I don't mind starting over too! Exciting times.
Am I the only one shaking my head that they kept comparing the M3 chips against M1 chips skipping M2?
Just like the last product release spent more time on huffing their own farts (the Apple Watch is the first of its kind carbon negative product), there is no real innovation or something to garner interest for me here.
Way more people bought in to M1 than M2, so they're a bigger customer base to sell to. But didn't I see something about M3 being 2x faster than M2? I'm sure that's some BS about it being 2x faster in a specific benchmark. But I'm sure it's still a good generational improvement.
Edit: I love seeing what HN will downvote. Absolutely hilarious.
Not real innovation (when comparing with Windows world) but having support for hardware raytracing now is definitely nice for games.
However, then on the information page they show Myst. Doesn't look spectacular to be honest, probably doesn't make use of hardware raytracing anyways. Having a short video of a fast-paced shooter with hardware ray-tracing support would be more impressive.
I guess it will take some time until games will adopt new Metal graphic APIs for hardware raytracing.
I understood they wanted a comparison to the M2, the most recent version. Most people care about “what changed since last version” it’s disingenuous to say “X faster!! (Than an model out date by a year)”
The focus seemed particularly on people still on intel macbooks and selling them on upgrading. I don't think they would usually reference a product that outdated at this point but they brought it up several times highlighting 11x faster.
I think it makes sense in that the people that bought M2 are not likely to be in the market for an M3. The people looking to buy an M3 are likely upgrading from either an M1 or an older Intel based MBP.
The mainstream marketing for iPhone 15 is only highlighting that's its made from titanium... I'm sure there is some innovation in manufacturing, and I do love my 15, but it's devoid of any innovation.
A few years ago, I tried to figure out the battery life of the latest iPhone. The comparison page would only say "2 more hours than iPhone X", "4 more hours than iPhone 8", etc. Eventually I worked my way backwards to the iPhone 6 where the battery life was just listed as "—".
The M2 was a tiny upgrade over the M1 since the 5nm to 4nm step was tiny, 4nm was basically 5nm. The M3 is probably running on TSMC's 3nm which is a huge improvement over both their 5nm and 4nm processes.
There's the obvious, that they want to say higher numbers.
But also, I bought a 16" MacBook Pro M1 Max when it was announced. The M2 Max replacement was announced only 15 months later and didn't seem like a big enough improvement to justify buying. Maybe they know there are a lot of M1 owners with my mindset, and we're who they're pitching.
In the same way that Wikipedians rush to update the pages of recently dead celebrities, I'm surprised there isn't a similar rush to update pages like https://en.wikipedia.org/wiki/Apple_silicon around events like this.
Sadly, only a handful of GPUs out there have hardware AV1 encoding chips. If I read the product page correctly, that new M3 architecture still cannot encode AV1 from hardware.
1,161 comments
[ 15.6 ms ] story [ 2106 ms ] threadThe presentation explicitly highlighted this very randomly, without showing how AI development actually works.
I know Apple has put more effort into Apple Silicon support for PyTorch but is it there yet?
See eg. https://www.reddit.com/r/LocalLLaMA/comments/16o4ka8/running...
M3 Max might be a contender, but Metal is far from CUDA yet? A laptop M3 Max looks like it can compete with a Nvidia 3070.
It's an unfair comparison as Nvidia is a huge desktop heat/energy hog and Apple M chips are really efficient. Let's see how desktop M3 chips fare.
A lot of things work out of the box with pytorch on mps, but not everything, and it can be frustrating to figure what works and what doesn't while you have a bunch of other moving parts shifting under your feet due to how fast the whole thing is moving.
That said, llama.cpp (aka "ggml") on M1/M2 Mac is very robust. If you just want to run inference (as opposed to training) of models and you're comfortable with a Mac, there's really no reason to go out and by an nVidia card.
FWIW, some numbers I found on reddit: https://www.reddit.com/r/LocalLLaMA/comments/16o4ka8/running...
Especially since (and this just occurred to me), in a roundabout way the original may have been the first Mac game to ever use raytracing (in its pre-rendering)
I was just coming here to ask about the Myst reference, since I played that 25 (?) years ago and wondered why they mentioned it. So there's a remake?
e.g. https://www.globaldata.com/data-insights/technology-media-an...
Tims a finance guy, they are kinda evil by default.
Even having to upgrade the RAM is a hassle when price comparing etc. They know this and it's - by design - misleading to the type of affluent less-tech-friendly users who prefer Apple products.
I mean, surely a mixed bag, and I still participate in it regardless, but pretty much, yeah.
https://www.reuters.com/article/us-usa-mining-children-trfn/...
https://www.culawreview.org/journal/child-labor-and-the-huma...
Im just some punk trying to have a good time with some spicy comments on HN. It's ok to both hate and exist within capitalism. It requires a certain level of pain (masochism) to survive that existence, or dare to enjoy it, while also raging against it.
Seriously, you can't make up this stuff..
And that $1599 USD is without taxes. So not the whole US has this price. It was always worth for me to travel to Oregon to have zero tax.
So I’d rather have to do some basic math but save 10% (of course different tax rates only partially explain the price difference, it still is significantly more expensive in Europe without VAT)
Perhaps what's ridiculous is that we have so many different taxing jurisdictions.
Why is that ridiculous? I don't pay sales tax, what else would you have them advertise?
- Macbook Pro (M3, 8GB RAM, 512 GB Storage): $1549 -> €2049 ($2179) = 40% Markup
- Dell XPS 15 (i3-13700H, RTX 4050, 16GB RAM, 512GB Storage): $1865 -> €1989 ($2110) = 13% Markup
- Framework 16 (Prebuilt Performance Pro - Ryzen 7840HS, 16GB RAM, 512GB Storage): $1739 -> €1907 ($2023) = 16% markup
Apple's EU markup is an outlier here.
Still, US sales tax is low. EU is >= ~20% almost everywhere.
Sadly (or perhaps gladly, depending on your perspective), there just is no equivalent to a macbook for many people.
I was a long-time Mac user. I remember when my 2006 MacBook cost less than comparable PC laptops. My last Mac purchase was a refurbished 2013 Mac Pro I purchased from Apple in 2017; since 2021 I've switched to PCs. While I love the configuration choices and expandability of PCs, the operating system situation leaves much to be desired, in my opinion. The shenanigans Microsoft continues to pull with Windows as well as the Sisyphean development cycle of the Linux desktop ecosystem (systemd, Wayland, etc.) makes me peek at Apple whenever there's an announcement, but the steep prices for a Mac configuration that fits my needs are too much for me, and so I remain a PC user.
Around 6 months ago I actually bought a beefy desktop PC to play around with, dual booting both Windows and Linux and they were just terrible. The default keyboard shortcuts on Windows are a joke (not to mention, the lack of an additional modifier key which is present with mac, which makes shortcuts more difficult than it needs to be).
Linux on the other hand, was so limited. People then argue that Linux is customizable, but I spent forever trying to customise it and half the time, the modifications literally don't work at all. I assume because there are so many distros and the eco-system is so fragmented that something is almost guaranteed to break.
I don't see how they're equivalent. F2 is vastly inferior to pressing the enter key to renaming a file. Alt + F4 to close a window as opposed to cmd + w.
Can you name shortcuts that are more convenient on Windows than on Mac?
That wasn't my argument
> They seem to be just as inconvenient and requiring a rebind on both systems
But they exist: Screenshot with PrtScn (or even Windows key + Shift + S) is better than the 3/4 Mac ones
> F2 is vastly inferior to pressing the enter key to renaming a file.
It's actually close to objectively better (though not a good one):
- Enter is a side-ways-stretch pinky key, so one of the worst keys out there (Backspace is worse since it's diagonal stretch)
- F2 is a middle-finger vertical key, so more convenient
Besides, this blocks opening files with Enter in the dumb Mac file manager, so add that to the downsides of superior Mac shortcuts
I think more than anything, it's difficult to discredit something that you don't actively use and aren't completely familiar with. Like I'm just saying Windows is shit because it's not Mac, but that's disingenuous because I don't actively use Windows as my main OS. But hey, if you like Windows, use Windows. If you like Mac, use Mac.
I think the base MBP from 2015 had 8GB.
I waited for today's announcement because I'm in the market for a laptop with 32GB of RAM. Today's announcement looked really tempting! However, $2399 + AppleCare + sales tax for the 36GB M3 Pro MacBook Pro is too much for me, and so it looks like it's either a refurbished M2, a Framework laptop, or a ThinkPad for me.
You just need patience and enjoy the fall. Just a few years and soon enough nobody is going to care developing stuff for their platform, it will be mostly empty shells, but at least they are pretty, I guess. Patience my friend, karma will do its thing eventually...
When I bought it, I knew it was going to be a laptop that I'll keep for at least 5 years. So far, it still feels like new.
On a separate note does anyone have any insight on how unified memory compares to vram in the context of machine learning performance? Considering an H100 with 80GB costs like $30k a maxed out macbook pro with 128gb unified memory for $5k is interesting. Is it remotely comparable or compelling for large models considering realistically most prosumers are capped at the 24gb nvidia cards for anything resembling a reasonable budget.
https://pr.tsmc.com/english/news/2986
Maybe not. It’s a bit of a pain having to unplug and replug everything in, not to mention having windows and such shuffle around each time.
I have no idea what Apple's sales are for the macbooks or if Airs sell more than Pros, etc., but they're definitely making profit this way to be this consistent in their approach.
Nvidia's approach doesn't have this.
1. Engineering bandwidth (it's work to upgrade a lineup! They probably need at least a year)
2. Manufacturing bandwidth - they probably spent the first half of the year manufacturing the 3nm iphone processor while researching the mbp processor. I expect it's difficult to ramp manufacturing on many chips simultaneously
3. Sure, consider the demand side. If my parents are going to buy a new laptop and a new phone this year, I think they would be more likely to do that 6 months apart. Similarly, I expect keeping a cadence of announce products a,b,c in this quarter and d,e,f in that quarter helps keep apple in the news (for good/exciting things)
Wrt machine learning: I can't wait for the results to come out for this once we can get our hands on it
Sort of. It's that the whole concept of a "ramp" implies that manufacturing things (not just chips!) starts expensive and risky and get cheaper and reliable over time as people work out the kinks in the various processes.
The M2 is mature. TSMC can make them reliably and in high quantity at low marginal price. Why would you *not* sell a product like that if there's a market for it?
Apple introduced the 15 inch MacBook Air on June 6th.
If they upgraded it to the M3 right now that machine would have only lasted 4 months and 4 days.
Sure they could do it, but how many sales are they really losing because they haven’t? People buying the Air are not exactly looking for the absolute best in technology. They’re probably much more likely to be price sensitive, and putting a brand new M3 in would likely make that worse.
So if you consider that you have to make new boxes and manuals and new hardware and test it and change the lines and everything else… what are the chances you would come out positive after such a short amount of time?
It would make some more sense for the 13 inch air which is about 18 months old, but again if you don’t have to maybe you keep selling the M2 for another six months.
I forget how many fabs TSMC has that have 3-nanometer capability, but it's gotta be low. Apple's supposedly got _all_ of TSMC's capacity[0], but I doubt it's enough.
[0]: Apple is saving “billions” on chips thanks to unique deal with TSMC | https://news.ycombinator.com/item?id=37040722
It costs less to manufacture the current generation ship and the last generation chip concurrently, it allows them to push the newest generation chip out the door faster without running out of supply or having to open up excessive production lines.
>H100 vs apple silicon
while it's only one spec, H100 has over 2TB/s of memory bandwidth and Apple silicon caps out at 400GB/s. H100 in general kicks the pants out of Apple silicon performance and has far more support. These parts aren't really in the same league but Apple Silicon has become a popular hobbyist choice for LLM inferencing.
Also see: The Jetson AGX Orin 64gb, Nvidia RTX A6000 48gb, AMD Radeon Pro W7900 48gb, two nvlinked 24gb 3090's.
For the M1 and M2 generations, the airs released in the fall and the pros a few months later in the spring. The airs have gotten so good that many people that would have gotten the more expensive (and more lucrative) pros got airs instead.
* ~~Ultra~~ Max probably has an adequate limit with "up to" 128GB
* Pro would be nicer to have "up to" (say) 64GB, rather than 'only' 36GB
* Base perhaps "up to" 32GB, rather than 24GB
There are some situations (running a VM or three with different OS(es)) where there really isn't a substitute for more bits.
But naturally, if you want to run on 128 cores and 4TB of ram, then you can’t do that in a laptop anyway.
Not surprised. But I agree with you. Especially for anything with the word “Pro” on it.
e.g. https://www.space.com/spacex-drone-ship-a-shortfall-of-gravi...
They don't "make" them, they design it.
I suspect any $2,600,000,000,000 company could pull it off, if they moved first.
https://en.wikipedia.org/wiki/P.A._Semi
https://en.wikipedia.org/wiki/Intrinsity
https://en.wikipedia.org/wiki/Apple_A6
Nope. Apple and AMD both use TSMC for manufacturing. It's all made by the same factory. AMD does not have the advantage there. Apple buys the most capacity on the most advanced process nodes since they place much bigger orders (Apple also has 10x more cash than AMD).
The real question is why Apple left it so long. They clearly wanted the first generation to be a clear success, but they could have probably pulled this off faster had they wanted to.
EDIT: Digging this a bit, it's not (one or more) 72-bit wide busses with 2^32 words as I'd expect as a gray-beard, it's (probably) six 32-bit wide busses with 6 GiB per bus; and this use of 1.5 * 2^N deep memories has become relatively common with the use of IC stacking, with 12 power-of-two sized ICs stacked in a single package (instead of the more "comfortable" 8 high or 16 high stacks of the same ICs giving powers of two).
Clive Sinclair did that trick back in the 80s.
Instead, it looks like after the 8-high stacks that I knew about, manufacturers went to 12-high stacks. There are 16-high stacks available too, but looks like there's a lot more backside thinning needed to keep the same Z height so cost goes up disproportionately and 12-highs are a reasonably sweet spot for capacity per dollar.
M2 and M3 go up to 24GB
M3 Pro is available in 18GB and 36GB
M3 Max is available in 36GB, 48GB, 64GB, 96GB, and 128GB
Rumor is Apple bought all of TSMC's initial 3nm production. They are probably yield limited on full featured chips.
Often, people will search for a random part that fulfills the same function, sort by cheapest, and then let their indignation run wild, completely ignoring the difference in form factor, other properties, or even quality.
That said, it’s no secret that Apple adds a healthy margin and an “inconvenience” tax.
In Apple’s ideal world, all people would purchase a handful of mass-produced configurations. This saves them in manufacturing costs, assembly costs, and logistical costs.
Apple also spent an ungodly amount on engineering to “make more with less”.
In the long run, this saves them money on lower-capacity components, especially at the quality and with the ancillary properties they're purchased. This is why spec for spec their iPhones and Macs look underpowered compared to competitors while performing the same if not better.
So, from their perspective, it’s “fair” to upcharge the “spec peepers” and professionals who really need it. The latter is generally less price-sensitive.
But even when one could just pop in your own RAM, many would still buy Apple's upgrade, which sounds insane on the surface. But there are actually pretty sound reasons for it:
1. They're buying the Mac hardware because it works. That's one of the core motivations for spending the extra to begin with.*
2. Adding ram is the gateway to unexpected crashes, such as errors that only pop up during heavy use, higher temperatures and the like. It doesn't have to be cheap either.
3. But if one is going for those cheaper components, one gets what they pay for: "mislabeled"(online fraud schemes), counterfeit, bottom-rung binned components sold as legit are par for the course.
--
* A lot of brands say their products work, when they simply don't. Here's some of my own examples:
I'm now onto my 3rd brand of mesh wifi routers - why? Because what's out there is garbage, even when you pay a lot for it. It's clear that there is insufficient QA on many top brands, and they simply won't acknowledge that the product they've sold you doesn't perform.
I purchased one of the pricy 5k LG screens that were a total lemon, at this level of expense you don't expect school-boy errors in hardware. LG handled it poorly, both in their return and exchange options, as well as customer support and patches.
Specifically, when it comes to routers, it seems that manufacturers only ensure that the basic routing works adequately and that all the bells and whistles exist solely for marketing reasons.
This is so bad that their CS is trained to have you disable said marketed bells and whistles during troubleshooting, only to conclude that everything is fine as long as the basic routing functionality works and that anything that doest work beyond that is a “you” problem.
3x6GB = 18GB
3x12GB = 36GB
3x32GB = 96GB
4x12GB = 48GB
4x16GB = 64GB
4x32GB = 128GB
That sounds like 3 memory controllers vs 4 memory controllers, likely due to chip binning. TSMC's 3nm processes supposedly still has relatively low yields.
I'm running the lowest spec m1 with 8 gigs of ram for my daily driver.
I frequently have tens of chrome browser tabs open across multiple profiles, multiple chrome debuggers, at least two or three VS code instances, multiple high memory nodejs processes, video conferencing, screen sharing and building code all at the same time - and the thing runs smoothly without a hiccup. And does it on battery! For a full work day, at least.
Weighs less than a pixie's fart.
I can't even imagine the capabilities the new m3 max has with that kind of memory and power available.
It's going to pinch, but I guess I'll be finding out in a few weeks.
My m1 is so good it seriously makes me wonder if I really need to upgrade.
What kind of workload are you running on yours?
What is your usecase, LLMs? Other AI models? Multiple large VMs? It’s always worth it to max the memory. In this case going from 64 to 96.
Their designs, being primarily mobile-focused, are rather hard-coded to specific memory configurations.
Or a 12 plus a binned 8?
Not sure why my initial comment is -4 votes right now. What am I remembering wrong?
"MacBook" (not Pro) was plastic from the start, and introduced the same year as the MacBook Pro (always aluminum): 2006. The plastic MacBook was discontinued after the 2010 model. There was a one-off aluminum 13" MacBook in 2008, but it coexisted with the plastic MacBook and was not replaced by a plastic model—instead, the 13" model was promoted up to the MacBook Pro product line, which previously had been just 15" and 17".
After several years of not having a plain MacBook, Apple did the 12" aluminum MacBook from 2015 to 2017. It was basically what they wished the MacBook Air could be, but was too expensive and too thermally constrained to truly replace the MacBook Air of the time. It was killed when they finally upgraded the MacBook Air to have a Retina display.
I don't think it was the Pro version (if it even existed). I certainly didn't have money to buy anything but the cheapest model.
Apple likes to wait until it can announce something with a most around it. In this case, more compute isn’t enough, I don’t think.
This has to be a product differentiation decision at this point, right? Is there any serious technical limitation against more displays in the base chip, that Apple has not been able to solve in three generations?
They do, you have to pay more for an M3 Pro or M3 Max.
That combo isn't currently offered, and judging from today's M3 specs the next MBA won't likely offer it either.
I assume the iMac has the same limitation, which is a bit of a downer for some uses. It would be nice to drive a second monitor and a wall-mounted TV or projector from an iMac in sleek a home setup.
...and almost any Windows laptop at the same price point.
I finally bit the bullet and bought a DisplayLink dock, which gets the job done, but just seems like a silly compromise when the Intel MBA could support 2 external monitors natively.
For me, still on m1 is good news, one more generation I can skip without feeling too much FOMO. Graphics improvement are irrelevant since im not gaming on them and cpu difference is fine for being just <40% overall.
How is that a downgrade?
WTH, was Apple sandbagging on the M1?
The performance improvements here is roughly what you would expect if you knew the specs of TSMC 3nm chips.
When the m1 came out details about the m3 were well known and that this was going to be a pretty big generation for performance improvements. I think the surprising thing was that m2 got pretty close to m3 in situations which were not as thermally constrained.
Nice! When can I order these most advanced PC chips for my next build? Do they have any kind of upgrade path for existing AMD/Intel users? I don't mind starting over too! Exciting times.
If you have several billion dollars to convince them to sell them for you as standalone chip, very soon!
Just like the last product release spent more time on huffing their own farts (the Apple Watch is the first of its kind carbon negative product), there is no real innovation or something to garner interest for me here.
Edit: I love seeing what HN will downvote. Absolutely hilarious.
However, then on the information page they show Myst. Doesn't look spectacular to be honest, probably doesn't make use of hardware raytracing anyways. Having a short video of a fast-paced shooter with hardware ray-tracing support would be more impressive.
I guess it will take some time until games will adopt new Metal graphic APIs for hardware raytracing.
Scroll down to "Power and Battery", about halfway down the page.
But also, I bought a 16" MacBook Pro M1 Max when it was announced. The M2 Max replacement was announced only 15 months later and didn't seem like a big enough improvement to justify buying. Maybe they know there are a lot of M1 owners with my mindset, and we're who they're pitching.
https://cdn.arstechnica.net/wp-content/uploads/2023/10/Scree...
It's taken years but AV1 has ended up everywhere.
Encode for h264 and HEVC was slow too. It basically never even happened for VP9 outside of phones, AFAIK.
https://i.postimg.cc/TYpx7SVf/Screen-Shot-2023-10-30-at-8-48...