I had to return a decently spec-ced M3 Macbook PRO 14" because it only supported 1 display (base M3) and pay more for M3 pro even though I don't need the extra horsepower.
And now the base M3 Air's support 2 displays? This is wild
where I measured tokens in my requests before sending things to openai API to ballpark the cost
There are other browsers that are far more memory efficient.
What about colima vm for docker? Intellij, gradle daemon, android studio, xcode, electron apps, android virtual device, slack, spring boot apps, iMovie, Davinci Resolve, python notebooks?
Phones now have more than 8gb of RAM
The limiting factor won't be the memory but the thermal throttling.
It is exactly why I haven't bought anything from Apple since the ipod.
Everything sounds amazing/wonderful and then one mind boggling deal breaker that doesn't even seem to make economic sense but just feels petty.
It’s incredibly stingy. For a while you could make the argument that a lot of Air purchasers wouldn’t need it, but I don’t think that’s the case any more. My wife has an M1 Air with 8GB and between Office, Chrome, Teams, and Slack she quite regularly gets beachballs and weird performance hitches.
Open 1000tabs and it will kill any machine.
In any case, it's pretty easy for the OS to swap out browser data as it is chunkable by tab. Just because it had allocated 22GB doesn't mean that it was all active. Must was certainly swapped to SSD.
The only application I ever need to terminate is Notes, which for some reason is extremely crash prone on the mac.
What is that?
I've been using macs for decades, and I've never seen or heard of such a dialog.
Never had it with Windows. Or iPad, or iPhone. Why does the pro computer do this? My guess is 5k external monitor and a few desktops, but...really. I'm not running multiple layers of VM's, it'll crash it with a few VS code windows, browsers, command lines.
Look, at least the Apple is smart enough to wake up basically in the same window state after having refreshed the resources. That's fantastic. But even better to fix it without going through the whole "have you turned it off and on again" routine.
Different models of computing. The iPhone only displays one tab inside one application at a time. And switching is slow enough that the OS can prefetch the swapped memory for the new view. A desktop app usually keeps everything alive as switching is very random and instantaneous. And you don't go open things that require a lot of RAM when you don't have that many (Each electron app is a whole new browser which is already a resource hog)
Considering how bad all of that is about "weird performance hitches" (read: running an entire browser for every single app) are you sure that has anything to do with the memory?
Those are like the heavyweights of the app world. And I think the 4 combined is running more code than the OS itself (the last three are 3xChrome).
Using Slack as a web app[1] solved my issues with it. It runs using Webkit instead of Electron, which is far smoother.
I looked at the price for m3 and if it would have 16gig it would literally be perfect for just 1k but nope 8 more gigs cost you an arm and a leg
I had a Macbook Pro 2015 with 256GB SSD. (Base model was 128GB). It was a very painful experience even back then. Yet almost ten years later, we are here, still paying $200 to upgrade to 512GB, when almost every Windows laptop comes with 512GB. FYI a 1TB NVMe gen 4 drive costs less than $100.
Compared to what? Linux? Windows?
Are there any published benchmarks anywhere, otherwise this just proves my point above.
> Apple tends to have faster RAM (way more bandwidth), faster SSDs
Yes, sure, due to its integrated nature, but that does not reduce the RAM requirement. My 8GB M1 MBA, which is used as a home browsing-only laptop, is almost always in yellow on memory pressure once we have a few tabs open.
Got it!
Anyway, sometimes people know what they're talking about.
The RAM or SSD are specifically not faster because of its “integrated nature”. The RAM is faster because Apple engineered an actually wide bus + multiple channels for memory access. The fatter Macs have the equivalent of up to 8-channel memory, which not even server CPUs of the competition provide. The SSD gets easily 7 GB/s reads in my testing. Both Windows and Mac have a working memory compression algorithm and a sister post claims that it works better due to hardware acceleration, which I’m inclined to believe. Memory compression on Linux is a mess as you cannot keep compressed memory pages in physical RAM, instead forcing you to partition the memory and manually handling how much and what to put in compressed zram.
Low memory situations are thus handled better than on other operating systems. Memory and memory pressure are neglected concepts in the competition, both the hardware providers and operating system providers.
8 channels has been standard in Epyc chips since they released in 2017, the latest Epyc Genoa does 12 channels. They're also not splitting that memory bandwidth with a bandwidth hungry GPU.
I am sick of people making claims without quoting any numbers about real-life performance.
Think of an alternate universe where Apple does the opposite: every new model they push the envelope and double the baseline RAM compared to the previous year. In that world you’d have all the software growing in memory use without bound. Consumers would be forced into a treadmill of computer upgrades like we haven’t seen since the 90’s when CPUs were skyrocketing in performance every year.
For anyone who forgets what the 90’s was like, here’s an example with Mac models:
1990 saw the launch of the Mac LC which had a 16 MHz Motorola 68020
1999 brought the Power Mac G4 at up to 500 MHz
That’s a 31-fold increase in clock rate (and several times that in overall performance) in the same timespan we’re discussing. Software that was written for the G4 had no chance of running on the LC (ignoring CPU architecture differences).
MacBook Airs are the mainline consumer machine these days. Apple does not want users to feel like they need to upgrade them every year (despite what people say).
If you are talking about "native" apps, maybe. Otherwise, nah. Cross-platform apps based on web like Teams and Spotify won't put too much effort on performance as long as it is not too slow. And if you haven't realized, most of the stuff you interact with is online. People just shove an entire website.
As for professional apps -- if you can't run a heavy audio/video editing application smoothly, I'm pretty sure that's your problem. Developers can put more effort into optimizing for 8GB RAM, but at the end of the day these workflows require large amount of memory, and after a certain point it is not worth to optimize for this segment of users
Shouldn't you be doing the opposite then? Keeping the baseline amount so you can know what it's like for people without a large budget and stop patronizing the applications without acceptable footprints in that circumstance?
> In that world you’d have all the software growing in memory use without bound.
We already live in that world. In the 90s you could run Netscape Navigator on a machine with 8 megabytes of memory. I've seen individual browser tabs use more gigabytes than that.
And not all of this is Electron bloat. The Stable Diffusion XL model is ~13GB. In general the quality is going to be proportional to the model size. So for the thing to get better, people need machines with more memory. And 8GB is already too small.
I'm not a developer of native Mac apps. If I were, I would definitely have a baseline machine for testing.
The Stable Diffusion XL model is ~13GB
That's not a baseline consumer application. See my other reply (re: grandma and little Billy). If you're developing a native Mac app for grandma and little Billy, Apple probably doesn't want you shipping a 13GB model with it. This is an example of the point I'm trying to make: find a way to compress the model so the end user doesn't have to deal with that kind of bloat (or host it in the cloud).
You expect every developer to buy a second Mac? They're all doing what you're doing and paying more for the machine with more memory because other applications need it, and then their application runs fine on that machine so they don't even notice the problem for the people with 8GB.
> That's not a baseline consumer application.
It will be before any of these new machines go out of support.
It's very easy to check the memory usage of your project (it's kinda in your face on XCode). If you do a few minutes test or let the app stay open for a few hours and usage has ballooned to a few GBs, that usually means you have a leak somewhere to fix.
The thing that really bothers me is that if you look at what regular home and office users were doing on their computers in the 90's it's almost identical to what those users are doing right now, except in the 90's they had orders of magnitude slower computers with orders of magnitude less memory. Yet in many cases those 90's computers were MORE RESPONSIVE than what they have today.
All of those countless billions of investment in technology hasn't done a damn thing for the productivity of Sally the office worker or Billy the 6th grader. Arguably, it's made their lives worse (viz. social media's deleterious effects on mental health). Now everyone's pushing the heck out of AI and all I see is high schoolers using ChatGPT to cheat themselves out of an education. They can't read (critically), they can't write, they can't even spell!
So in light of all that, why should we be pushing more and more computing power (and memory, the original issue) on regular users who aren't getting any benefit (broadly, to their way of life) out of it?
Gosh, now I sound like a luddite!
Certainly not, but you can't fix it by putting less RAM in the machines of people with budget constraints. The developers will just pay for more themselves and then not care about those people because people who can't afford RAM generally aren't lucrative customers.
And it's also worth considering what actually causes this.
Developers want their code to work on every platform. They don't want to write different code for each platform. But each platform wants them to have to, because that makes it more likely there will be software that only works on their platform, or that doesn't work on some new competing platform. So they refuse to develop or implement cross-platform standards.
Then someone else has to do it, but that's rather a lot of work, and it turns out the easiest way to do it is to piggyback on the work already done for browsers to make them work on every platform. That's Electron. It's terribly inefficient but it saves the developer a lot of porting work, so it's widely used.
If Apple doesn't like this, they should provide cross-platform native APIs for developing applications.
It’s not in Apple’s interest to do that. It would cost a lot of money to develop and only benefit the competition. It would also slow down Apple’s own ability to innovate on the APIs until the competitors catch up.
Or are you saying Apple should develop the APIs for Windows and Linux as well? Why would they do that?
What they should do is provide open source implementations of their APIs for Windows and Linux, i.e. make them standards.
> Why would they do that?
Because then people would use them instead of using Electron.
Nah, the treadmill stalled out years ago.
You're also using a time period that includes COVID and when the DRAM manufacturers got busted for price fixing.
That's not a run of the mill consumer set up.
>You're also using a time period that includes COVID and when the DRAM manufacturers got busted for price fixing
No, I'm actually including two separate time periods where they were busted for price fixing. They were doing it in the late '90s too.
It kind of is, e.g. Steam hardware survey has more than 80% of people with at least 16GB of RAM and almost a third with 32GB, and that's a measure of installed base rather than new computers.
You can now find 64GB in <$550 laptops:
https://www.newegg.com/p/1TS-000D-110E6?Item=9SIA7ABJ459240
> No, I'm actually including two separate time periods where they were busted for price fixing. They were doing it in the late '90s too.
They started doing it at the end of the '90s. By then PowerBooks were already coming with 64MB and PowerMacs with 128MB:
https://everymac.com/systems/apple/powerbook_g3/specs/powerb...
https://everymac.com/systems/apple/powermac_g3/specs/powerma...
In the second case it started only 3 years into your measurement period instead of 7, and then right after that was COVID. It's only now that the prices are starting to resume their historical downward trend and they're still slightly above where they were when the price fixing started in 2016.
But that explains why it's not a factor of 64 during this period. It's still the case that 8GB of DDR5 is ~$24. What reason is there to not include $100 worth on a $1000 machine? Or, if some excuse for that could be generated, why isn't there a $1100 machine with four times as much?
You're saying "RAM usage", but your evidence is "RAM provisioning", which is the entire basis for the criticism. The reality is that RAM usage has increased significantly, but Apple has been stuck by the addressable memory limits they've baked into their architecture.
As another point of comparison, a 2013 iPhone 5c (technically not released until the last quarter of 2013, but we're being generous here) had 1GB of RAM and 8GB of storage, though you could upgrade that to 32GB. A modern iPhone has 6GB of RAM, and comes with storage from 128GB to 1TB.
Maybe the bytes are growing more on iOS? ;-)
Fine, take that as your reference. It still shows that growth in memory was 1000% faster not too long ago.
...and that still proves nothing about the growth of RAM usage in that time period, because the size of the offerings in individual products are largely independent of the increases in RAM usage.
I've yet to notice the impact on getting web sites to stop using incredibly bloated JavaScript that leak memory, video conferencing & streaming apps from using codecs that redline the CPU, or game developers from writing games that make the GPU cry uncle, or...
> 1990 saw the launch of the Mac LC which had a 16 MHz Motorola 68020.
You got lucky, because we got the Mac II Si back then, but they were both kneecapped on the factory floor by the 16-bit memory bus that crippled the 68020's 32-bit memory bus. 2 year old PCs ran circles around it. Planned obsolescence was was one of Apple's crowning achievements back then.
> 1999 brought the Power Mac G4 at up to 500 MHz
The Macbook Air from ten years ago would be Retina I had: a 2-core i7 that could go up to 3.5 GHz. Ask me how well that runs software written for the new M3's. ;-)
> MacBook Airs are the mainline consumer machine these days. Apple does not want users to feel like they need to upgrade them every year (despite what people say).
You might have missed this bit from Apple's blurb on the new MacBook Airs: "13x faster than the fastest Intel-based MacBook Air". The fastest Intel-based Macbook Air was produced [checks notes], 4 years ago. It's hard not to read that like they aren't trying to convey a need to upgrade.
If you're wondering why people are saying what they're saying, it's because Apple is saying what they're saying.
But at the same time, I've never seen non-techies complain that they can't open a few tabs, reply to a few emails, watch youtube, listen to music, and study on their base laptops. They're amazing for that.
If we need more memory to accomplish the above tasks any time soon, we're in a sad state of software development and bloat.
Again, I'm personally not happy about shelling out $400 for 24GB of memory on my next Macbook, but I gave away one of my base machines with 8GB to a family member, and they simply did not notice.
A web browser adds a significant amount of memory usage to any task, even with basic static webpages, and then almost every other app you use is secretly also a web browser but none of the RAM usage from it can be shared with the web browser you're already running.
The vast majority of their market would not fit into the developer, content creator, or just plain power-user category. There's people who just log on to do internet banking and email when they're not on Reddit or Facebook.
For some anecdata, 16/512+ would be a waste on my parents, in-laws and my whole extended family for that matter. They would benefit from it, but they're not screaming at spinning wheels, and are probably a bit more patient and accustomed to 'slowness', which is pretty subjective.
Even worse, if one plots the (price, unified memory amount, chip type), and looks at it from right-to-left, then the dollars per system capability is disgusting when you order a lesser system. You get the most value for your money when you buy the maximum unified memory configuration (on those three points).
Better yet, with a maxed out unified memory configuration, one can further save on SSD writes by using (and loading/storing) RAM disks for their projects!
Most people buying a MacBook for work are likely getting a higher unified memory, so their workstations live longer. Meanwhile consumers will have to keep consuming as they fry their internal SSD's on their airs...
I was also wondering if a 1TB apple fabric device is simply an 8TB fabric device with 8x the write life...
Have a crate with about 40+ of those machines at work, essentially useless from the era where the person in charge of buying laptops just bought the bottom end for every employee and considered it a job done.
I have a great M2 pro machine, but officially it can only support 2 external monitors. I should be able to close my screen and power 3. I can do this with a dock so it's not a resources problem.
I am curious what is different between the Air and the Pro that the Air can power 2 external monitors (it does say when the lid is closed) and the Pro can only power 1 regardless. Or is this a software update and the Pro page just has not been updated.
I keep hoping that this is a problem that is only temporary and eventually it will be removed or as time goes on each series can run 1 more monitor or something.
A TB4 dock fixes most problems, but I have to hack the EDID to disable YCbCr and force RGB, or the colors look like absolute shit. The external monitors still look significantly worse under macOS than either Windows or Linux, and I have no idea why.
External display handling is easily the worst part of using a Mac for me.
Using more than is explicitly mentioned that is supported, you mean? I mean yeah, duh.
I've been running three for a while. 2 Thunderbolt 3 + 1 Thunderbolt 2 daisy chained. They work. There are a couple glitches, but they work.
I'm on my work machine now but I believe it also works with the laptop open.
https://9to5mac.com/2024/03/04/14-inch-m3-macbook-pro-multi-...
But nevertheless, Apple's hardware strategy sucks ...
[1] https://www.bestbuy.com/site/samsung-odyssey-49-1000r-curved...
[2] https://old.reddit.com/r/ultrawidemasterrace/comments/mb9vnv...
Considering minor Macbook upgrades can get you into >$1500 price territory pretty easily I think this a fantastic value. If you wanted to buy my full setup right now it would be $849 + $749 + $145 = $1745 but you're getting “dual” monitors and dock that can be reused with any modern machine, making it easy to switch between work and play. I can even plug my Steam Deck into it. :) (No affiliation with any of these products!)
One thing you should be aware of is that having two seperate monitors can be an advantage for Window management on macOS. With two monitors, I can swap spaces on one of them and keep everything as is on the other. With only one it's not that easy.
I use Asahi Linux on M2 now, and the USBC display support isn’t done yet, but I am hoping it would be better than MacOS.
More-than-one-display is pretty much the standard on regular (non-apple) computers: you can drive as much as your system can sustain.
If it can identify something people will pay more for, while not quite putting off most people, it'll do it, no matter how mad.
So if most people think 8GB is fine for a laptop, they'll buy it, and everyone else pays through the nose for more.
Most people only want one extra screen? That's what you get on the base machine and you must pay (alot) for more. If one day most people need two, then you'll get two.
Apple seem to spend a lot of design time on how to extract the most money out of those willing to pay, no matter how annoying it is.
I don't want to waste time migrating to a new system, only to find out I need to return/resell it, or later down the road find out another arbitrary/artificial limitation Apple has set that I either have to find a work around for or suck up until I can switch machines again if none exist.
This is unfortunate since there are some features that have made it very tempting to switch.
But just in general, Intel has always prioritized lots of I/O flexibility on its chips. If you look at the datasheets there have always been dozens and dozens of units on every SKU that are never plumbed out to ports on the device. Three or four display outputs, six or eight USB controllers, stuff like that. Apple is the opposite: they won't include something if they aren't absolutely sure they need it. So after the shift from x86 to Apple silicon, laptop users are feeling a squeeze on I/O that used to seem "free".
The M3 Pro has more space and is more ... pro?
Apart from the initial M1 MacBook Pro release, it feels like most products Apple has released in the last few years has always been missing one or two features, and the next release happens to have that feature. E.g. the first M1 Air did not have MagSafe even though the Pros did, and then Apple included MagSafe in M2 Air, but it didn't support multiple displays; now Apple is including multiple displays in M3 Air.
It feels awfully convenient that each generation conveniently has a nontrivial feature upgrade.... Apple has less incentive to make each generation "complete" -- by delaying features (more) consumers will feel obligated to upgrade per generation.
Edit: Better wording, I suppose, that the non-PRO supports two external monitors with the lid closed, the PRO supports 1. Still an odd overall offering/branding.
I had a 2019 cheesegrater Mac Pro. With Catalina, I was able to drive two 4K screens at 144Hz in HDR10, because Catalina apparently supported DSC 1.4.
Then they introduced the ProDisplay XDR with Big Sur which had people agog at "how were they able to drive this 6K display given the bandwidth limitations?"
Well, the answer is because they absolutely nerfed/bastardized DSC 1.4 from Big Sur (and it's maybe only been updated in Monterey? Unsure, I no longer have the screens - ironically I bought an XDR) to make it happen with some proprietary magic: those same screens could now only be driven at 60Hz in HDR10 or 95Hz in SDR.
Proof in the pudding was that my monitors (LG27GN950-B) actually allowed you to change the advertised/supported DSC version, and when I "downgraded" the monitors to DSC 1.2, performance actually improved, and allowed 120Hz SDR and 95Hz HDR.
This happened with many many users, across many screen types.
Apple studiously ignored it, and may still be. They simply don't care if you're not using an Apple display.