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.
M2 to M3 vs M3 to M4
- Thread starter macrumor2018
- Start date
- Sort by reaction score
From what I looked up, I don’t think either is a significant difference. I know M3 introduced some power saving capabilities that M2 didn’t have.
As to future proofing, I think that’s more hype than reality. If you buy a system that has the spec you’ll need for what you’re doing currently and likely in the future the processor isn’t going to make that much of a difference. I don’t see someone in the future saying if I had an M4 instead of an M3 I could keep this Mac for another five years. It’s going to last longer because it’s one newer generation, but it’s going to last one generation longer. Nothing more.
I would like to hear some expert opinions if they think one is a bigger leap in technology
As to future proofing, I think that’s more hype than reality. If you buy a system that has the spec you’ll need for what you’re doing currently and likely in the future the processor isn’t going to make that much of a difference. I don’t see someone in the future saying if I had an M4 instead of an M3 I could keep this Mac for another five years. It’s going to last longer because it’s one newer generation, but it’s going to last one generation longer. Nothing more.
I would like to hear some expert opinions if they think one is a bigger leap in technology
The following results came from ChatGPT with "Web search" and "Think" enabled. Take with a grain of salt. 
For CPU, M2 -> M3 has the biggest generational leap. The process node upgrade (from 5nm to 3nm) should be the biggest contributor.
* These figures are approximate “relative-to‑M1” values aggregated from benchmark reports and announcements. Variants (Pro/Max) show further improvements.
For CPU, M2 -> M3 has the biggest generational leap. The process node upgrade (from 5nm to 3nm) should be the biggest contributor.
| Metric | M1 (2020) | M2 (2022) | M3 (2023) | M4 (2024) |
|---|---|---|---|---|
| Process Node | 5 nm | 5 nm | 3 nm (N3B) | 3 nm (N3E) |
| Transistor Count | ~16 billion | ~20 billion | 25 billion | 28 billion |
| Single‑Core Performance | Baseline (1.0x) | ~1.15x improvement* | ~1.5x improvement* | ~1.65x improvement* |
| Multi‑Core Performance | Baseline (1.0x) | ~1.17x improvement* | ~1.5x improvement* (varies) | ~1.7x (base) – up to 2.1x* (Max) |
| GPU Performance | Baseline (1.0x) | ~1.41x improvement* | ~1.45x improvement* | ~1.75x improvement* |
| Neural Engine | ~11 TOPS | ~15 TOPS (approx.) | ~18 TOPS | ~38 TOPS |
Approximate Generational Gap (Relative Improvement vs. Previous Generation)
- M1 → M2:
– CPU: Approximately +12–16% improvement in single‑core performance and around +16% in multi‑core performance.
– GPU: Roughly +41% improvement relative to the M1. - M2 → M3:
– CPU: About a +30% jump in single‑core performance and roughly +28% in multi‑core performance over the M2. - M3 → M4:
– CPU: Roughly +10% additional gain in single‑core performance (bringing M4 to about 1.65x over M1) and multi‑core gains vary from around +10% on the base model up to as much as +30% for Pro/Max variants.
– Neural Engine: Notably, the M4’s Neural Engine nearly doubles (from ~18 TOPS in the M3 to ~38 TOPS), representing a ~100% increase.
M3 brought hardware raytracing to Apple Silicon, and while I'm of the opinion that it's mostly a gimmick when it comes to gaming, I understand there are some 3D workflows that take good advantage of it.
Regardless of my opinion on future proofing and I'm not in total disagreement there, the M3 was a huge change in capabilities from the M2, especially in the GPU: dynamic cache, ray tracing, mesh shaders, etc ... And the CPU for both the M3 and M4 were substantial upgrades in performance and architecture (SME). To give you an example using Cinebench R24, for the combined M3/M4 generational increase going from the 8+4 M2 Pro to the 8+4 M4 Pro, the latter is 43% faster in ST while being 22% more efficient as well - while in multicore the M4 Pro is 36% faster and 34% more efficient and no the improvement in TSMC fabrication nodes don't account for those kinds of changes.
So from my perspective, capabilities wise M3 introduced the most new functionality, but it can be argued that the M4 had the bigger leap in performance and introduced some new stuff like SME matrix calculations for the CPU (though the AMX matrix accelerators existed in previous generations of Apple's CPU, you could only access them through high level frameworks instead of ARM-approved CPU intrinsics which limited their use cases).
This sounds interesting, but how does it translate to English? I’m being serious here because all these tech specs and benchmarks are great but what does it mean when the average MacBook user is at his computer? I think that would be helpful because I suspect most people don’t know what an SME matrix is. How does it relate to what I’m doing on my Mac? I feel like people get dazzled with fancy words that they don’t understand so they just upgrade.
I know everyone talks about video editing, but how many people out there with MacBooks are doing video editing? Maybe it’s more than I’m thinking?
I think capabilities and speed for work would be more understandable. Maybe some of the new artificial intelligence capabilities and how they help workflow? It would be nice to see examples of people’s daily use on these different processors and how they are different or improved. I know there has to be improvement, but it would be nice to hear a real world. Examples of how someone’s day is made better by going to M4.
It depends ... the original user asked about specs, features, and future proofing. What you're asking is different, how important are those to user X? If you are day-to-day user of Apple Silicon, happy with a base M1 checking email, web browsing, and light document work, then probably the most impactful thing you would notice is slightly increased responsiveness and a better battery life/performance on the more intense tasks you do happen to do from time to time. General performance and efficiency improvements help everyone but indeed how much you notice will depend on what you're doing and the rest may not mean much. So if you aren't interested in pushing your computer in gaming, video editing, rendering, AI, development, scientific computation, etc ... , then no I guess you wouldn't notice most of these things. But if you ARE interested in any of the above then yes, you'll notice. SME and SSVE are used for low power AI and scientific workloads (maybe a couple of others, but those are the main two I can think of). Meanwhile, dynamic cache, ray tracing, and mesh shaders offer massive improvements to gaming, game development, and 3D rendering in general. In aggregate, how many people are doing task Y on an Apple computer?, it begins to add up to quite a substantial number of people. Further those people drive the sales to other people since they create the market for apps in the first place and of course include the developers for said apps.
Also, I disagree with this notion that average consumers "just upgrade" because they're told fancy words so they ditch their old computer. Yes there are enthusiasts who upgrade early and often because they enjoy having the latest and greatest tech (and of course professionals in any of the aforementioned fields tend to upgrade more often), but that's not the same thing. Even a lot of the average lifecycle of a computer being 4 years is driven by business cycles. No one who casually spends $1000+ every couple of years on an Apple computer just because they hear a buzzword is, in my book, an average consumer, not even an average Apple consumer.
It's not a gimmick unfortunately, new AAA like the last Assassin's Creed have RT baked in. This is why a baseline M3 beats an M1 Pro or Max in their minimum specs.
Between this and other functionalities like SME, SSVE, dynamic cache etc... I tend to agree that the M3 is the most significant upgrade and will age significantly better than M1 and M2, esp in GPU-heavy workloads.
M3-M4 was huge.
Why?
M3 was a little bit of a downgrade vs. M2 in a number of areas. It was still faster overall but gave up a bunch of memory bandwidth. M4 returned bandwidth to better than M2 levels.
M3 isn't bad. But its clear apple had to make some compromises to deal with the manufacturing problems the process had, and as a result it is less than it could/should have been.
edit:
to be clear, I'm talking about the FAMILY of chips, not m3 base vs. m4 base, etc.
Base chips - m3 was fine. but there was no m3 ultra, and the m3 pro spec was.... not great.
I think there are three product tiers emerging for Apple silicon. There is overlap between the high-end of the first two tiers and the entry-level of the last two tiers. Looking at the current N3E generation:
[1] A18, A18 Pro, M4
[2] M4, M4 Pro, M4 Max
[3] M4 Max, M4 Ultra, M4 Ultra+
I think each of these tiers is a little different. Apple has 15 years of experience with the first one, iPhone and iPad, beginning with the A4 in March 2010. Each generation features more than just new silicon, there are various changes to the chassis as well.
I think the second tier is now pretty much set. The main difference in marketing dynamic from the first tier is that the chassis doesn’t change with every generation. So there is an additional layer, the cadence for the chassis, that runs beneath the cadence for the silicon. That is still playing out, obviously, but it is dependent on the same kinds of things that shape the chassis on the first tier. Ultimately, I think many of the lessons with regard to consumer (average or otherwise) behavior Apple has learned from its long experience with the first tier also apply to the second tier.
The third tier is different. The silicon is far more important. I think Apple’s approach here is informed by its experience with Intel. These customers and their workloads demand tangible improvements in efficiency and performance. Someone with an M2 Ultra will upgrade immediately to an M4 Ultra (or beyond) because the machines pay for themselves, but if that equation doesn’t work, Apple’s got a problem. I don’t know, but I think it’s less about cadence and more about what they can deliver.
Last edited:
Wow - you can't even turn it off? It's cool to see the reflections in the water in CP2077 - it definitely adds to the realism - but it's certainly not groundbreaking. I'd rather go without and have the extra frames if the hardware is struggling.
More and more games will start using RT, as if it works it enables the developer to massively simplify the way they do artwork whilst giving a more realistic end result.
Traditional 3d rasterisation is a massive bunch of hacks to make something that looks "close enough" at significant cost to the art department. It's objectively pretty bad, but its been all hardware has been capable of in real time until recently.
Yes, just like in Star Wars Outlaws, you can't turn it off. Without hardware RT on M1 and M2 it defaults to software RT. Which is why Ubisoft expects M3 to perform better than the M1 Max.
Ah I see. I didn't think of it from the standpoint of being easier on the developer. It's just a bummer that a forced transition is already happening - it's gonna really ratchet up the GPU requirements across the board. I was hoping to get another couple of years out of my 4070!