Developer logs reveal more details about next-gen Apple M3 and M3 Max chips
- Thread starter JournalBot
- Start date
I'm using an M1 mini as my daily driver, which is an outstanding machine, but I, and my wallet, are extremely ready for an M3 Pro mini. It'll be a beast.
Upvote
8
(9
/
-1)
This is likely the biggest piece.
Apple can afford to pre-pay for orders. TSMC can then use those pre-payments to buy equipment, develop prototypes, and whatever else they need to do... WITHOUT needing to spend their own cash reserves, cut budgets elsewhere to free up free-cash-flow, or deal with taking out loans.
Not in the chip industry, but if a customer came to me asking to toll manufacture something for them (What TSMC does for Apple), I'd definitely let those customers willing and able to pre-pay for everything have first access to my facilities and their output. It's the old adage, "a bird in the hand is worth two in the bush". except the Apples business is 2 birds in the hand vs everyone else's 1 in the bush.
Upvote
35
(36
/
-1)
this is because apple's phones are recession-proof, and now that they have so much tie-in crap (think watches, which require an iPhone), its even more recession-proof than before!
the rest of the tech industry has the same downturns, ant that is why a big fab would have nearly flat profits
Upvote
-5
(1
/
-6)
It is not the Storage that gets you. Because I am more than happy with 512GB, and there multiple viable and reasonable paths to get more. (USB, Thunderbolt, cloud)
Now, Apple would say that storage is faster. Yet, I have a Frameworks with a faster SSD that is socketed than my M2-Max Apple Laptop. So I don't buy the reason I can't just have more storage. Especially considering the Mac Studio and Mac Pro.
What gets you is the RAM. if you keep your computer for 7 years, you have to predict what you will need in 2029.
As it is, I am in a Mac for the battery life. I would probably just stay with what I have. But the option to have socketed RAM would be nice.
Upvote
1
(10
/
-9)
Personally, and this may well reveal my age, the Apple processor is the most existing technological milestone of our era, because it would open the door to so much power with little consumption.
RISC & CISC architecture has reached its limit over two decades ago and we knew that when a fridge became required to cool it during operation. It is simply unsustainable and unpromising.
RISC & CISC architecture has reached its limit over two decades ago and we knew that when a fridge became required to cool it during operation. It is simply unsustainable and unpromising.
Upvote
-11
(9
/
-20)
Here's one site where you can do a side-by-side specification analysis (click the link to see the comparison chart):
https://everymac.com/ultimate-mac-c...icon005&prod2=MacStudio005&prod3=MacStudio006
If you're planning to keep the machine for more than about three years (which is probably most Mac buyers), I would recommend the Studio over the Mini because it has more ports and thermal headroom for sustained loads. But one main advantage is it can be configured with more memory (at a cost) which is an important decision for a machine with non-upgradable RAM.
Upvote
11
(12
/
-1)
"Just buy a new mac" and we'll throw the old one into a grinder to "recycle" it.
Upvote
-16
(2
/
-18)
Yeah new processes are becoming increasingly expensive. The machines needed for the processes. The clean room space. The development effort is all going up substantially. TSMC needs Apple's boat loads of cash that is very directly contributing to these continued advancements to stay on top. If Apple did leave TSCM for Samsung of Intel it's quite likely that TSMC would suffer major delays in their development of newer processes and have a hard time increasing fab capacity for cutting edge nodes.
Upvote
10
(11
/
-1)
Luckily the new Studio wasn't out when I bought my Mini, so I wasn't tempted. The M2 Pro works great for my workflows (my heaviest usage is probably Logic Pro, and I've never maxed it out, which used to happen with my Intel iMac). I don't think I would notice any improvement with the Studio. But YMMV.
Upvote
6
(6
/
0)
Not sure why you put “recycle“ in quotes. Yes, I know this is marketing copy below, but what Apple says on the page for new MacBook Pros is:
There are lots of conspiracy kook social media posts saying “Come on, we know they don’t really recycle, they landfill it” but I have yet to see any real proof that the above statements are wrong.
Upvote
39
(41
/
-2)
It doesn't necessarily give a higher profit as the process is tuned, since some tuning curve would be assumed in how both sides were negotiating the deal. I see it more as TSMC gets higher profit if their yield curve over time is better than expected, and worse profit over time if their yield curve is worse than expected (as compared to the traditional pricing model). There is likely some yield curve where both parties come out in the same position as a traditional pricing model would have left them. The deal seems to transfer more risk of underperformance as well as more return from over performance to TSMC, compared to the traditional deal.
Upvote
12
(12
/
0)
I tend to think of repairability and upgradeability as only somewhat related features. For instance, I honestly can't imagine the situation where I'd want to upgrade the screen on a laptop, but can very much imagine the situation where being able to repair a screen easily would be advantageous to me. An easily replaceable battery on a phone would allow for both repair and upgrading, on the other hand, although still I would suspect most people would never end up using that feature for upgrading.
Upvote
13
(14
/
-1)
Thanks. This is really helpful. I'm especially liking the Q&A segment for the Mac Studio as it does have a direct comparison to the m2 pro. Do you know of any resources for looking at them from the viewpoint of a developer? I don't do multimedia stuff at all, but I am a software engineer. So, those comparisons will make way more sense for me.
Upvote
0
(0
/
0)
These assertions are contradictory. “Perfectly good” and “no longer use” is on you.
My Intel Macs are all relatively old, dating back to 2016. They’re getting security updates and still do computer things just as well as they did when new. My M1 MBP is my new daily driver, but I could use the 2016 i9MBP just as easily, if with less speed and overall snappiness.
Upvote
0
(10
/
-10)
You might try poking around on the Ars forums to see if others have weighed in on this:
https://meincmagazine.com/civis/
If you're looking for the lowest-cost unit for your specific needs then the Mini is the best bet in the Mac ecosystem. The upper-end M-series Minis really are capable machines (unlike some of the Intel Minis in the mid-2000s). But if you really want something for the longer haul then bumping up a Studio's options should serve for the better part of a decade. Primarily, in my opinion, in terms of memory.
Depending on how many years you use it, the added cost might only work out to a hundred or so dollars more per year.
Upvote
5
(5
/
0)
One other comment on price comparison among Mac models, for those not familiar with Apple's site - I often open two windows side-by-side from the Apple Store:
https://www.apple.com/shop/buy-mac/mac-mini
https://www.apple.com/shop/buy-mac/mac-studio
(as an example)
Select a model from each and then choose the most common specs between two models. Sometimes there is a dependency on processor level for things like memory capacity (you'd need to do some trial-and-error selection to discover this).
If you can get two different Mac models specced out similarly (which these days pretty much boils down to CPU core count, GPU core count, memory, and storage) then you can see what other features differ, such as number of ports or Ethernet speed, and use those differences to see if the remainder of the cost delta can be justified.
This is a standard product comparison procedure, and for Apple products the prices at the online store give a pretty straightforward view of how the differences stack up. I usually shop this way, pick the configuration I want, and then wait 6-8 months to buy it as a refurb
.
https://www.apple.com/shop/buy-mac/mac-mini
https://www.apple.com/shop/buy-mac/mac-studio
(as an example)
Select a model from each and then choose the most common specs between two models. Sometimes there is a dependency on processor level for things like memory capacity (you'd need to do some trial-and-error selection to discover this).
If you can get two different Mac models specced out similarly (which these days pretty much boils down to CPU core count, GPU core count, memory, and storage) then you can see what other features differ, such as number of ports or Ethernet speed, and use those differences to see if the remainder of the cost delta can be justified.
This is a standard product comparison procedure, and for Apple products the prices at the online store give a pretty straightforward view of how the differences stack up. I usually shop this way, pick the configuration I want, and then wait 6-8 months to buy it as a refurb
.
Upvote
9
(9
/
0)
I am inclined to agree with you. But then I am not a gamer.
To me, Apple essentially is in the process of telling anyone with 32 bit programs to take a hike just like 16 bit and 8 bit programs before.
Not having to support a long tail of legacy stuff pays real dividends.
Intel can’t tell me that they made the perfect processor in 1981 and that all future chips need to be based on the x86 platform forever.
Upvote
10
(15
/
-5)
There are three basic issues with this.
1.) the unified memory architecture would need to be adjusted to include the possibility of the GPU for graphics. That is an architecture change.
2. Driver support. Someone has to write the drivers and get apple to include them in the OS.
3. Is Nvidia seriously in the GPU business anymore? I mean obviously they make GPU cards. But given the state of the 4xxx series and what Nvidia calls an upgrade, I am not so sure that they coincided themselves a GPU company.
All of these would combine for Apple to lead to dependency on an outside vendor for top end performance. And it sure looks like apple wants to develop their own path.
Upvote
9
(11
/
-2)
Folks who are closer to this than I am, a little help please?
I get that changes to the expected mix of instructions might change priorities for queue depth, energy budgets for different types of instructions, etc. All sorts of changes could cascade from a small tweak
But would they be the sort of “architectural improvements” that suggests there's some significant smarts that designers left on the table for the many years Apple's been working on its own silicon? Changes that they're only now getting around to putting in play? That just doesn't sound right for mature instruction sets
I get that changes to the expected mix of instructions might change priorities for queue depth, energy budgets for different types of instructions, etc. All sorts of changes could cascade from a small tweak
But would they be the sort of “architectural improvements” that suggests there's some significant smarts that designers left on the table for the many years Apple's been working on its own silicon? Changes that they're only now getting around to putting in play? That just doesn't sound right for mature instruction sets
Upvote
-1
(1
/
-2)
I have an LG monitor and an Apple Studio display connected to a MBP. The LG monitor takes a noticeable amount of time to wake up. The Studio monitor is awake instantly.
Upvote
10
(10
/
0)
Dude! You had a perfect opportunity to quote Dolly Parton's "Nine to Five", and you blew it!
And more to the point: aluminium is incredibly energy intensive to refine. It takes electrolysis to get the metal from the oxide; a friend once described the metal as solidified electricity, and he's not really wrong.: it takes about 17 kWh of electricity to produce one kilogram of aluminium metal. Recycling aluminium is much more energy efficient; there's a reason why recycling aluminium cans has been a thing for as long as I can remember (I'm now in my late forties), whilst other materials are johnny-come-latelies.
Gold is self explanatory. Rare earths... look, the material isn't exactly scarce, but the production of it isn't that high (especially if you want a China-free supply chain, which I'll acknowledge isn't Apple's primary concern at the moment), so there's a motivation to recycle that material as well.
So the simple economic fact of the matter is, there's money in it for Apple to recycle these systems as much as possible, rather than sending them to landfill. That on its own is enough for me to think that the burden of proof is on those claiming recycling isn't happening, rather than those claiming it is.
Upvote
23
(25
/
-2)
OrangeCream
Ars Legatus Legionis
Hold on.
Say you have a repairable Mac.
What do you do with the old RAM, SSD, power supply, CPU, GPU, case fans, WiFi module, motherboard, and HSF when they all get replaced or upgraded?
Fundamentally the problem is the same, right? With Apple’s current setup all of them get replaced all at once while in a repairable setup they get replaced one at a time over several years.
In the end they both need to be recycled, right? The key difference is that with Apple the CPU, GPU, and RAM are one unit 1/100th the size of separate CPU, GPU, and RAM, and the resulting HSF is also 1/10th the size of separate CPU/GPU/RAM.
I understand that you can increase longevity of a system if pieces could be replaced individually, but the best I can envision is SSD on a daughtercard, CPU, GPU, and RAM on a separate daughtercard, and WiFi, eth, and USB on a third, all wired together using some kind of PCIe bus.
Upvote
14
(19
/
-5)
One correction to the article: the A15/M2 is built on TSMC N5P, not N4P.
The A16 is built on N4P, but it is not the basis for M2.
The A16 is built on N4P, but it is not the basis for M2.
Upvote
5
(5
/
0)
If you have the space for the Studio with the M2 Max processor, get it. You get more gpu cores for your money with the Studio, a default 10GigE port plus ports on the front. I purchased the mini with the M2 Pro not long after it came out and the comparison was with the Studio M1 Max (I wanted a silent computer - the reports of fan noise were also an influence). My work is cpu based, so I didn't care about the gpu penalty - I was after the best cpu performance, so got the M2 Pro and maxed out the RAM (PyCharm can be a memory hog!).
Today my choice would be the Studio with M2 Max as the fan noise has apparently been addressed.
Upvote
7
(7
/
0)
I've replaced memory, screens, even upgraded the CPU daughter card. debugged a bad line on the floppy cables.
Just as I once built an MG from 3 wrecks—every system except the master brake cylinder & the rack & pinion; body work, paint—but no longer change my own oil, it no longer makes sense to invite the failure points of socketed memory & CPUs; no sense in acquiring repair skills & tools; risking errors or simply destabilizing it by losing screws, etc
Yes, it'd be nice to hold off on buying more RAM until my workload made it obvious I needed it, but net-net, it seems more time- and $-efficient to just over-spec by X% up front. Yes, it's nice to understand how your machine works but the mechanics seem a small fraction of what matters
PS: when my wife needed a new iMac years ago, the sales rep DOWN-sold her to a slower, lower-RAM machine. Served her perfectly well for a dozen years. The migration app wasn't perfectly easy but close enough
Upvote
9
(12
/
-3)
I'm with you on the exciting part.
My M2 Air is the first laptop after decades of them where I didn't feel like it was a compromise / I don't ever think about the battery.
Upvote
11
(12
/
-1)
I ran a Mac as my main system for quite a few years, and this became a source of frustration. Apple broke compatibility with my favorite programs time after time. There was always a transition period to soften the blow. The new OS will still run programs from the old OS—for a while. The new ISA will still run programs from the old ISA—for a while. Then eventually they don't work anymore. It wasn't the only factor that nudged me away from Mac, but it was a big one.
Of course, now I'm on the other side of the fence with my Linux-based system, gazing longingly at the Apple Silicon and all the performance and efficiency they're getting out of it, while I'm stuck with clunky old X86 seemingly from now until doomsday.
Upvote
6
(8
/
-2)
OrangeCream
Ars Legatus Legionis
I'm not 'closer', but I like to pretend I understand a little bit
Why do you think that CPUs can't improve?
Take the 2013 A7, a 64 bit ARMv8 design; why would you have assumed in 2013 that there weren't any future improvements available to the CPU? To be explicit, it was a 28nm part and a dual core 6 wide machine with 192 ROB, 1MB L2 and 4MB L3 and a die size of 102mm2
The A14 has 6 cores split 2/4 between BIG.little; it's an 8 wide/3 wide design, with the BIG part having a 630+ ROB, 8MB L2 and 16MB LLC on a much smaller 88mm2 die
Imagine if Apple is willing to take the extra transistors available with the shrink from 5nm to 3nm and grow the die size back to 100mm2; they would have, roughly, 2x the transistor budget over the A14. There might be areas before where they could get 80% of the maximum performance using 20% of the transistors, but if you could double the transistors you might get 90% of the performance (or a 12% improvement)
Or maybe they can revisit SMT, meaning that a core can process two threads instead of one. Normally you expect each thread to perform at half the level compared to dedicating one thread per core, but due to scheduling optimizations where the core can switch threads when a thread stall occurs you might actually see a net increase in overall performance of 12%
On top of that they might go from an 8 wide design to a 10 wide design; theoretically a 25% boost in IPC, but realistically a 0% boost due to limitations in thread level parallelism. However when paired with SMT you see a significant boost because now you can keep the pipeline more full. So maybe you can actually net a significant 20% increase in IPC (when paired with SMT)
Or maybe they go from 8 wide to 12 wide (50% increase in dispatched instructions) because their SMT implementation actually juggles three threads instead of two. Having the ability to draw instructions from three threads means they might actually see a much larger increase in IPC, so now instead of 20% increase you see a 30% increase in overall performance.
Or maybe they allocate the extra transistors to the little cores; instead of being a 3 wide design it grows to a 4 wide design, and with a bunch of tweaks that allow it to see, effectively, a 25% increase in IPC; going from 3 wide to 4 wide is 33% increase in dispatched and retired instructions.
Or maybe they go even wider and their little cores are now 5 wide with a corresponding 40% increase in IPC. They could also throw in a handful of more cores; so instead of boosting performance by making the BIG cores wider and then tacking on SMT, they make the little cores wider so that more threads can be serviced simultaneously. BIG cores might not see as much, but overall performance goes up because threads that previously had to be time sliced on a BIG core might now get more dedicated execution time on a little core.
Again, apologies if I've made basic mistakes, I'm not asserting any of these things will happen, or that any of these numbers are real. I'm saying that with a doubled transistor budget that Apple can revisit changes that previously were ignored due to cost per area implementations.
Upvote
18
(18
/
0)
Weird, my Late 2012 27" iMac still runs great! I use it when I work from home, and I haven't had any problems. I get that you probably using different apps and whatnot, but I paid a pretty penny for this machine and have no complaints (especially after upgrading the RAM from 8 to 24 GB).
Granted, I can't play Baldur's Gate 3, but oh well.
Upvote
5
(5
/
0)
Apple is too expensive you say?
Please do tell us more about this original thought you have come up with, that nobody has ever posted on Ars before!
Upvote
19
(22
/
-3)
I'm not sure that Apple could replace TSMC. TSMC remains significantly ahead of all their well-funded competitors. Building a fab company requires a lot more than just money; just ask China how their domestic fab industry is going.
If it could be done, it would take a decade and 10s of billions of dollars. Neither company can part ways from the other without massive negative impacts in the medium term. I suspect that whatever deal was struck was more about preserving a good relationship for the sake of future business rather than any realistic threats of Apple going anywhere else.
Upvote
6
(8
/
-2)
OrangeCream
Ars Legatus Legionis
Just like TSMC can’t replace Apple.
Yeah, so?
Where is TSMC going to get $16b? It’s a mutually beneficial arrangement and no one is ‘threatened’ by the other, though Apple more than others is often reported as a particularly hard bargainer.
Upvote
10
(12
/
-2)
The machine no longer getting fixes for OS security issues is a big problem. Unless this machine isn't connected to a network.
Upvote
8
(10
/
-2)
Not that every LG monitor will behave the same, but I'm disappointed to read this one data point. The next monitor I was considering was an LG, since they make the screens for iMacs (or at least they have in the past) and get decent reviews. I'm not sure what makes one monitor wake quickly and another slowly (I have my hunches but they're not based on fact), but unless Apple screws up their own designs they will likely have the fastest wake-response time.
(That was even demonstrated in a WWDC video with the opening of a laptop lid.)
That's why I'm inclined to get a 32" iMac if/when it comes out, assuming the rest of the specs meet my needs. Based on past experience I see myself getting seven years out of it, if not longer.
Upvote
-2
(0
/
-2)