So what is the Min RAM for each? 
Really hoping Apple moves off of the 8GB for the base M3.
Really hoping Apple moves off of the 8GB for the base M3.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
There is a table showing whether a feature is included in Armv7, Armv8 or Armv9.
A-Profile Architecture
The Arm Application-profile (A-profile) architecture targets high-performance markets, such as PC, mobile, gaming, and enterprise.
developer.arm.com
Arm's documentation is a bit confusing. Unless it is an optional extension, Mx has to have it because it is Armv8.5 compatible with some v8.6 extensions.
There is a table showing whether a feature is included in Armv7, Armv8 or Armv9.
A-Profile Architecture
The Arm Application-profile (A-profile) architecture targets high-performance markets, such as PC, mobile, gaming, and enterprise.
Arm's documentation is a bit confusing. Unless it is an optional extension, Mx has to have it because it is Armv8.5 compatible with some v8.6 extensions.
I think if you tell Arm that you are certifying for example 8.2 you can have 8.5 features in there as a incomplete 'next level'. You get graded on 8.2 and they can check the incomplete parts but not deliver a higher certification level.
Apple is pretty likely to 'cherry pick' the stuff they do and don't want to do. For example they have their own AMX for matrix. Decent chance the Arm matrix features will be a relatively very prolonged adoption. And that dance with Nvidia trying to acquire Arm; that probably didn't help rapid adoption either.
A perspective that I don't see yet in the thread is that the M-series of chips are direct offspring of the phone SoC architecture. This makes perfect sense for Apple as it makes efficient use of design effort, and provide a consistent platform across product lines. As the phones sell in vastly greater quantities than Macs, it seems reasonable to assume that the evolution of the M-SoCs will track that of the A line.
And the keyword for the A series of SoCs is efficiency.
This has served the Mac products very well. The efficiency inherited from the phone SoCs have enabled performant computers with excellent ergonomics and long battery lives, and the Apple silicon Macs have been extremely successful by Apple standards.
Apple is unlikely to change a winning (and sensible) concept. Thus they are unlikely to incorporate hardware that would be sitting around unused on the vast majority of phones - it's simply wasteful and stupid. Which in turn means that looking at technology trends in servers, Linux/Windows computational workstations or lunatic fringe gaming systems for inspiration for upcoming M-series SoCs constitutes approaching the question of future M-series SoC designs from the wrong direction.
Just tossing out bigger numbers and current tech world buzz-words is pointless without motivating exactly why it would be useful to Apple and Apple phones in particular. Remember, if Apple did absolutely nothing architecturally in the move to TSMC 3N, and just used it to lower power draw and reduce die sizes, that might actually be the advance that the bulk of their customers would appreciate most.
And the keyword for the A series of SoCs is efficiency.
This has served the Mac products very well. The efficiency inherited from the phone SoCs have enabled performant computers with excellent ergonomics and long battery lives, and the Apple silicon Macs have been extremely successful by Apple standards.
Apple is unlikely to change a winning (and sensible) concept. Thus they are unlikely to incorporate hardware that would be sitting around unused on the vast majority of phones - it's simply wasteful and stupid. Which in turn means that looking at technology trends in servers, Linux/Windows computational workstations or lunatic fringe gaming systems for inspiration for upcoming M-series SoCs constitutes approaching the question of future M-series SoC designs from the wrong direction.
Just tossing out bigger numbers and current tech world buzz-words is pointless without motivating exactly why it would be useful to Apple and Apple phones in particular. Remember, if Apple did absolutely nothing architecturally in the move to TSMC 3N, and just used it to lower power draw and reduce die sizes, that might actually be the advance that the bulk of their customers would appreciate most.
The most vocal of minorities demand a jump to i9 & 4090 raw performance with the M2 Ultra.
The performance trajectory of Apple Silicon shows it will eventually occur with each process node improvement and microarchitecture.
Chicken vs the egg... the hardware has to be there to take on the software.
Apples primary interest for hardware RT is production renderers, where they have already invested a large amount of resources. Other applications (such as games) will surely come. In fact, Apple has a real chance of being the first vendor to offer good RT performance on baseline devices, making RT applications attractive across the entire ecosystem.
Imagine 6+ million M3 Macs with good RT performance next year. It isn't much but it will get people talking about it.
Once Apple transitions fully to 3nm Mac chips then all 28+ million/year of them will have it by default.
I am not sure that a computer with 8 GB of RAM can provide a good ray tracing experience.
If you are using 8 GB base model for ray tracing, the joke is on you. I can see Apple offering RT only in pro/max/ultra.
I was talking about baseline performance, not experience. Obviously if you are trying to render a huge scene with 8GB RAM your experience is going to suck. But it all depends on what you want to do.
They will offer it across the entire line, from the iPhone and up. Apple does not differentiate by feature, they differentiate by capability.
Why is this a waste of time? Then again, back in the day some people called programmable shading a waste of time…
Extrapolation is dangerous. There are two major issues with this graph.The most vocal of minorities demand a jump to i9 & 4090 raw performance with the M2 Ultra.
The performance trajectory of Apple Silicon shows it will eventually occur with each process node improvement and microarchitecture.
One is that it stays within the specific period where intel was stuck at Skylake iterations and 14nm (+++++). Neither is valid anymore, although it is anybody's guess just how well Pat Gelsingers roadmaps will pan out, since we are looking forward to Raptor Lake refresh on intel 7N (old 10nm), and the promised Meteor Lake is still nowhere to be found alongside with the relatively straightforward 4N process node....
The other is that we know, (and I feel that this is something that there is really hard resistance to accepting) what future lithographic processes will bring - which is very little density wise and even less performance wise.
Anton Shilov at Anandtech has produced a digestible summary article comparing TSMC nodes, and an update three weeks ago for more specifics on projected backside power delivery timeline (end 26/beginning 27) and benefits.
What TSMC has shared tells us that we can look forward to hopefully 60-70% in total density increase the next 5-6 years including the benefits of bpd at the tail end of the period.
Those are, historically speaking, miniscule advances. Thus, emphasizing efficiency is more important than ever, and clever software being the ticket to performance and functionality advancement. Potentially backed of course by judiciously selected hardware feature enhancements, but I'd contend that unless these make sense for the 300 million or so annual A-series SoC customers, we are less likely to see them, as it would in a few years be dead weight added to a billion or so A-series SoCs. Thus I also believe that this autumns 3N iPhone SoCs will give us most of the answers to the M3 questions.
Yeah, I really wish people would stop throwing these kinds of graphs around and pretend like the trend is obvious. Not to mention that this specific graph is highly manipulative as it omits A15 and A16 where the trend is already levelling off.
My #1 wish for the M3 chips is for the base M3 to support more than 1 external display. It's so ridiculous to have to step up to the Macbook Pro to simply use two external displays.
We have a lot of users with laptops and docks at work, and it's crazy to me that an otherwise commodity-grade PC laptop that costs ~$500 can drive two 4k displays through one Type C port like it's absolutely nothing. Adding insult to injury, I'm extremely vexed that using the MBA in clamshell mode doesn't net you an additional external display.
The Macbook Air 15" would've been the ideal system for me but the lack of multi-display support ruled out that option.
We have a lot of users with laptops and docks at work, and it's crazy to me that an otherwise commodity-grade PC laptop that costs ~$500 can drive two 4k displays through one Type C port like it's absolutely nothing. Adding insult to injury, I'm extremely vexed that using the MBA in clamshell mode doesn't net you an additional external display.
The Macbook Air 15" would've been the ideal system for me but the lack of multi-display support ruled out that option.
Ray tracing is the standard in offline rendering and having ray tracing hardware would speed up it in macOS.
Could Apple think that ray tracing is currently only useful for 3D rendering and just add hardware accelerated ray tracing to M3 Pro/Max/Ultra?
Last edited:
Chiplets, I'm expecting the M3 to use a chiplet design.
The base M3 will look a lot like the M2 but with improvements here and there like folks have talked about here, but with an external cache chiplet. For the pro and above, on the desktops I'm imagining that you can order them with extra GPU chiplets.
The base M3 will look a lot like the M2 but with improvements here and there like folks have talked about here, but with an external cache chiplet. For the pro and above, on the desktops I'm imagining that you can order them with extra GPU chiplets.
I think this is fairly unlikely, as it would require them to develop and maintain two separate GPU designs simultaneously.
Raytracing often uses less memory than traditional rasterisation.
If you want to rasterise a decent sphere, you'll probably need at least 10k triangles.
If you want to raytrace a sphere you need 16 bytes.
Then if we're talking about shadows, you might need an entire cascade of shadow maps, while with raytracing you just need to cast a single ray towards the shadow-casting light and see if it hits anything.