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.
M4+ Chip Generation - Speculation Megathread [MERGED]
- Thread starter GoetzPhil
- Start date
- Sort by reaction score
previous ipad pro had m2 without hw ray tracing, so no apples to apple comparation
Here its another story
Today's announcements were interesting! Of course I wanted more from the A18, but what we actually got was very much what I'd expected, with a few interesting exceptions.
RT perf up 2x is a significant improvement, and I think we'll see that in the rest of the M4s. You may object that that's not in the base M4, but Apple's already shown a willingness to have different blocks in different models of the same generation (for example, different E cores in the M3 Max compared to M3/M3 Pro).
Other than that, the GPUshows no significant performance improvement- it's supposedly 20% faster than the A17 Pro, but it has 20% more resources (6 cores instead of 5). [Edit: I've been reminded that the A17 had 6 cores too. So this is a meaningful improvement!] However it's not clear if efficiency has improved beyond node improvements (N3B->N3E). They are clearly much more focused on sustained performance now (which I think is a good thing!!) and that would be a really significant change, which would be greatly appreciated in the coming M4 Macs as well.
CPU is a bit of a mystery so far. The improvements, while not huge, are again substantially more than you get just from the node switch. I don't think the P cores in the M4 could help there, as their IPC improvement isn't enough - their big advantage was scaling efficiency better with higher clocks. So it's possible this is another modest generational IPC improvement. It's not coming from clocks as they're boasting power improvements as well. It could possibly be coming form better E cores as well. I'd expect any such improvements to be in future M4 Macs as well.
There are a pile of other changes... I need to see if their tech specs page has more info. I doubt there will be any other useful sources of info for the next week or two.
RT perf up 2x is a significant improvement, and I think we'll see that in the rest of the M4s. You may object that that's not in the base M4, but Apple's already shown a willingness to have different blocks in different models of the same generation (for example, different E cores in the M3 Max compared to M3/M3 Pro).
Other than that, the GPU
CPU is a bit of a mystery so far. The improvements, while not huge, are again substantially more than you get just from the node switch. I don't think the P cores in the M4 could help there, as their IPC improvement isn't enough - their big advantage was scaling efficiency better with higher clocks. So it's possible this is another modest generational IPC improvement. It's not coming from clocks as they're boasting power improvements as well. It could possibly be coming form better E cores as well. I'd expect any such improvements to be in future M4 Macs as well.
There are a pile of other changes... I need to see if their tech specs page has more info. I doubt there will be any other useful sources of info for the next week or two.
Last edited:
So it does! Then that's a substantial improvement!
I'm still more interested in the sustained performance improvements, but this looks pretty good for the coming M4s.
Oh yes, one thing that went by pretty quickly - it looks like they now officially support access to their matrix hardware without going through Accelerate. Though it's possible that I misunderstood that. That's less critical for the Mac than the iPhone since Mac software doesn't have to pass App Store review. But it probably means that we can expect the instruction set, registers, etc. not to change (though it may grow), which is significant, and there will probably be a lot more support material made available over time.
Damnit, I also forgot to write about memory bandwidth improvements. I'm pretty sure that this is just an echo of what we've already seen with the M4 in iPP - higher clocked LPDDR5, in this iPhone at 3750MHz - so not indicative of anything for the coming M4 Macs that we didn't already know from the M4 iPad.
1. The M4 I-pad pro, that is also A 18 pro ? There has not changed anything, same "tech" ?
2. so we effectivly can expect the M4 to be 20% faster on Singlecore speed than the M3 ?
3. in total probably up to 40% better than the M2 ?....lets say with a slight increase in clockspeed vs. the phone or Ipad
4. and since they lower the power drain, there is effectivly room for a higher clock, say on a M4 studio ?
2. so we effectivly can expect the M4 to be 20% faster on Singlecore speed than the M3 ?
3. in total probably up to 40% better than the M2 ?....lets say with a slight increase in clockspeed vs. the phone or Ipad
4. and since they lower the power drain, there is effectivly room for a higher clock, say on a M4 studio ?
One thing beyond the ray tracing cores which we'll have to see how that compares to the base M4 is that the GPU clocks didn't improve from the A15 to A17. For the A17 to A18, same as the M3 to M4, there's a ~10% clock bump helped by the node transitions since the A15/M2 from N5P to N3E.
Last edited:
Do we know that? Surely it could be in the base M4? It just wasn't mentioned for iPad Pro because they were talking about improvements with regard to M2, not A17/M3, as @MayaUser has already pointed out...
Last edited:
They did say the M4 GPU was "based on" the M3 so it's not like they omitted the M3 entirely from their discussion, they just didn't elaborate on that very much and it seemed to be just a clock bump.
I've checked Solar Bay. Max Tech says the 9 core Solar Bay M4 was 58 FPS and Shrout says the 10-core M3 got 51 FPS. Unfortunately different core counts but assuming the 10 core M4 would get about 63 FPS that is substantially more than the 10% clocks alone would expect (24% improvement) and it's possible that while Solar Bay has RT, depending on how much RT affects performance in that particular benchmark, that 2x might not show up as substantially as it would otherwise in a different, more RT-heavy benchmark? (Not sure about this, it has been contended that Solar Bay has light RT). So it's possible these new RT cores are in the M4? But it's also possible that they are new. Doesn't really confirm one way or the other. We'll have to see what Apple says in October/November unless someone gets a more definitive answer first!
I'm strongly leaning towards these new RT cores being already in the base M4 and Apple just ... didn't mention it. But I'm far from certain.
Last edited:
Lol at the subpar GPU upgrade 18 Pro vs 17 Pro. I actually believe Apple might hold back on the A chips now because they know if we find out M5 is gonna have a huge GPU upgrade, we won’t buy upcoming M4.
Do we know yet how they doubled RT performance (or what RT "feature" they are talking about being doubled if not the whole "process")?
GitHub - tzakharko/m4-sme-exploration: Exploring the scalable matrix extension of the Apple M4 processor
Exploring the scalable matrix extension of the Apple M4 processor - tzakharko/m4-sme-exploration
github.com
- A18 Pro single-core score (one result): 3,409 single-core score (+18%)
- A17 Pro single-core score (average): 2,896 single-core score
- A18 Pro multi-core score (one result): 8,492 multi-core score (+18%)
- A17 Pro multi-core score (average): 7,192 multi-core score
The M4 runs at 4.4 GHz and the A18 at 4.0 GHz.
Neither their wording nor the benchmark results suggest that A18 brings any improvements over the M4 cores. The GB6 scores are consistent with clock increase, and the biggest driver of score improvements is the new SME matrix ISA, just as what we saw with M4. The only notable difference is the 2x RT performance claim, however some are arguing that it was already present in M4 based on the stinger than expected improvements in the Solar Bay scores.
We would need to look at the power figures for more info. M4 was drawing more power than ever at single-core peak. If Apple has managed to optimize this with A18, that would be a very meaningful improvement.
It’s SME, which is already available with M4.
Very unlikely:
1) That's not their usual behavior. They tend to roll out things as they're ready. This doesn't always apply to entire products - witness the skip of the Watch Ultra 3 this year - but it has applied to their chips.
2) They are very clearly NOT introducing all new features in annual lockstep any more. For a perfect example, look at the difference between the E cores on the M3 Max versus the M3 and M3 Pro.
So if they introduce new M4Pro/Max next month (they could just call them M5, if they felt like it), it's easily possible - even likely, I'd say - they will have some differences in some cores, compared to the M4 in the iPP. If they have a better GPU now than when the M4 shipped, it'll be in the Pro and Max.
BTW, I'm not sure why you'd say that it's a subpar upgrade. Seems pretty strong to me.
Last edited:
Yes, I think now that it's more likely you're right about the P cores. GPUs remain to be seen, but certainly possible as well.
As for SME... sigh. My dumbest moment this year? I plead lack of sleep...
I think I should clarify what I meant.
I mean that since 6 months have passed since M4 was introduced, they could have put their latest GPU tech (or latest at the moment 18 Pro was finalized, which I, incorrectly perhaps, assume was ~6 months after M4 was.
And that would have led to higher gains for the 18 Pro, if it used "M5" graphics cores. But instead it shows roughly the same improvement as the M4 over M3, right?
So what I meant was this may have been deliberate, because if there IS a strong GPU improvement bound for M5, and they let 18 Pro get that, well we would have found out, and may have decided to hold off buying M4.
I find it interesting how Apple has been increasing the frequency of its SoCs for iPhones and Macs in recent years.
| A18 Pro: 4.04GHz | M3 Max: 4.05GHz |
| A17 Pro: 3.8GHz | M2 Max: 3.7GHz |
| A16 skip | |
| A15: 3.2GHz | M1 Max: 3.2GHz |