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
Interestingly only the top end M2 Max/Ultra was ~3.7. The cutdown Max/Ultra was the same lower frequency as the Pro/base, ~3.5. Unclear why Apple chose to do that and they didn't exactly make a big deal out of it - probably an experiment. But Vadim's more or less right with the progression.
Last edited:
Man I just can't wait for this stupid phone hype cycle to be over, because I'm ready to buy a new Mac... M4 Max time!
Yeah, I’m really hyped by the M4 Pro and Max, and very intrigued about the upcoming M5 which will likely come in the N3E process as well… maybe they’ll focus on the GPU for the M5 gen? Or faster RAM? Or perhaps a Neural Engine with double the cores (32)?
Yah the M5 should be very cool! But I won't own an M5 device because I'm ready to buy now. I'm expecting this to be a really nice chip, mostly because of N3E!
Not reasonable. N3E is a modest improvement over N3B. At least half, and probably the majority, of whatever advantage you wind up seeing will be due to more advanced and/or larger design, not process.
It's possible, even likely, that the M5 will be on the next node, N3P, given the timings. After all N3E went into volume production in the second half of last year and the M4 on N3E was released in May of this year.
You’re right! I had totally forgotten about the N3P process node.
The node we need for the true desktop/workstation Macs (Mac Studio/Mac Pro Cube/Mac Pro) is N3X...! ;^p
Aye, for the M2, but Apple didn't do it in the M1 or M3 generation and didn't really talk about it. So we don't have a very good sense of why Apple experimented with binning for clocks this singular (so far) time. It may just have been an experiment on their part, which they like to do - they're constantly fiddling with the E-core/cluster, especially on the Max, and it's possible based on Gurman's missives that the entire M3 Pro die will be a singular experiment and they'll go back to the M4 Pro being a chopped die of the Max. But who knows, maybe they'll go back to binning clocks for the M4 Max or maybe for Hidra if Gurman's leaked information that it is a desktop-die are correct? Truthfully, the boost wasn't massively impactful, like 6%, so it isn't like they went full Intel/AMD turbo boost XXX and 10 different chip lines based on the same die.
Apple's strategy so far seems to be to set clocks low enough to harvest as many working die as possible and do one bin mostly by core count. This works in part because they create a relatively narrow range of devices and don't sell to OEMs who are each looking to create a hundred different products to differentiate themselves from their competitors. In contrast, AMD/Intel/Qualcomm are trying to salvage as many die as possible by stratifying on both core count and clocks that they can sell to said OEMs each looking to create that "unique" product. So it's just a different approach. It has to be said that Apple's strategy, at least wrt the CPU, also works in part because they have such nice a microarchitecture lead that they don't have to stratify by clocks to get really great ST performance. Depending on how much pressure they get put on in the near future that may change of course, though hopefully we won't see them go down the same rabbit hole as AMD/Intel if it does.
Last edited:
As you guessed, your initial assumptions about A18 Pro are almost certainly incorrect, in two ways. First, keep in mind that A18 appears to be a binned A18 Pro. They are not different designs, unlike (for example) M3 and M3 Pro.
The second is the production timeline. Keep in mind that Apple will need something like 150 million A18 in the first year of sales. That’s roughly 10x the volume of M4 needed. Volume sorting and testing of A18 would have begun at the same time as M4, or even before M4, building inventory for months prior to assembly ahead of the launch.
That's what I would have expected, and it may well be true, but there are some oddities that don't quite match up with what you'd expect if this were the case. Apple very specifically said *and displayed in text* at the keynote that the A18Pro has more cache than the A18. That's a pretty strange thing to bin on! It's not impossible, but it would be very new and different. So while the odds I think are in favor of you being right, I'm not ready to just assume that and move on. Hopefully we'll eventually see die shots, or someone with some clever microarchitectural benchmarks will come up with a more definitive discrepancy.
If they’re actually binning them, I think it’s much more likely it’s because the plain A18 has one less GPU. But even based on GPU, that’s a relatively small fraction of the die area.
There is the possibility that they’re just artificially limiting the A18 chips from A18pro production. Nothing wrong with that. It is a very common process, especially if binning does not provide sufficient quantities of chips
Agreed, they could and that would be fine. But again... fusing off some cache would be a VERY weird choice. Fusing off a GPU or CPU core is a very common choice, and they've done that a bunch of times. This, not so much. Or I should say, never at all, and I'm not aware of any other CPU maker doing that either (fusing off cache).
I'm not saying I think they're different designs. I'm saying, I don't think we know for sure they're the same design, yet.
They would also need to disable the programmable machine learning accelerators in the A18 Pro CPU cluster, the supposedly better media engine and USB3 controller.
Maaaaybe. The USB3 controller behavior would be a product of how the port is wired up. No need to fuse anything on the chip for that. The rest could be done in the OS, rather than in hardware. But... Why would they do that?? There's no benefit to Apple from that. So I think your larger point stands, which is that this supports the notion that the chips really are different.
The better media engine is definitely a difference between the two chips, they mention that specifically. But whereas immediately before talking about the programmable machine learning accelerators in the A18 Pro CPU cluster they specifically differentiate the A18 and A18 Pro based on cache size whereas they just say the "new CPU has new machine learning accelerators that are directly programmable", they pointedly don't say that the A18 doesn't have that. It's almost certainly a reference to Apple's latest CPU matrix accelerator which have been in all Apple A and M chips for a while now and recently got the SME upgrade in the M4. My suspicion is that's in both actually but they wanted to highlight it in the Pro model (like with the ray tracing). Yours and @Confused-User's larger point still stands, it's not clear that the A18 is a binned A18Pro, though it could be! I just wanted to say that not everything that they only mentioned in the A18 Pro might be unique on die to the Pro as opposed to the things they explicitly said differentiate the two. (And even then as @Confused-User said while I was typing this response, some of that might not even be a die difference or by the same token represent a fusing off, it depends!)
Exactly.
Nvidia does on L2 for GPUs. Here are two AD103-based chips:
NVIDIA GeForce RTX 4080 Specs
NVIDIA AD103, 2505 MHz, 9728 Cores, 304 TMUs, 112 ROPs, 16384 MB GDDR6X, 1400 MHz, 256 bit
The 4080 has 64MB of L2 while the 4070TI has only 48MB. (L1 per SM is the same)
I know you said CPU maker but since the Apple SOC serves both the CPU and GPU (and NPU) and wrt increased cache in the Pro model I suspect they are referring to SLC rather than the CPU's L1/L2, it's within the realm of possibility. Again, very unclear if the A18 is a binned Pro or its own different design. I'm leaning towards the latter, but it's very slight and it could easily be a binned model. If/when we get 3rd party die shots, it should be more clear.
Last edited:
Wait, SME is an A18 Pro-only feature? That would mean that the base A18 in not M4-gen, and it would surprise me a lot.
I don't think it is unique to the Pro. For instance, they also only mention the new ray tracing core on the Pro model, but I'd be shocked if it wasn't in the A18 too and they don't explicitly say the new ray tracing cores aren't in the base A18. Similarly, as I wrote above, SME isn't explicitly mentioned as being an A18 Pro feature unlike the larger caches, GPU core, and some of the media engine features. My suspicion is that SME just isn't talked about in the A18 chip presentation and is talked about in the Pro chip presentation but it is present in both.
If the psychologists in the marketing department say that a few more differences between the product lines are likely to increase revenue through increased upselling then that's what Apple will do.
I don't know why they would highlight it specifically on the A18 Pro but not for the A18. That just caught my attention during the presentation.
Apple mention that they jump two full generations ahead in the event from the A16 Bionic to the A18, which they say is "created specifically for the iPhone 16" (timecode 44:52) but then moments later in the presentation Apple says the "A18 has our latest generation 6-core CPU" (timecode 45:40).
Sri mentions that "just like A18 is key to enabling iPhone 16, iPhone 16 Pro's unique performance and feature requirements demanded a new Pro chip. This is A18 Pro. It's even faster and more efficient than A18, and it enables many of the features that are unique to iPhone 16 Pro" (timecode 1:16:56). At 1:19:02 he says "compared to A18, it's backed by larger caches for better performance in the most demanding apps and games. This is the fastest CPU in any smartphone. The new CPU also has next-generation ML accelerators that help power Apple Intelligence".
I forgot about the new advanced display engine, video encoder, image signal processor and faster USB3 speeds mentioned for the A18 Pro at 1.19.31.
They are clearly two distinct SoCs and not just binned for two SKUs. Or Apple certainly wants us to believe that.
Last edited:
Honestly, I'm not sure. But possibly the same reason they didn't mention the ray tracing cores in the A18 or M4. The base A18 chip presentation was very short - they may have just saved more of the details for the A18 Pro presentation, perhaps as @Basic75 said because they want to push that model and make it seem more impressive. But as you quote yourself below Apple says that the A18 has the latest generation CPU and since the SME capable matrix accelerator is part of the CPU, so the A18 almost certainly has it.
Yup two separate sentences and you'll notice the graphics next to the cache says "compared to A18" while the next-gen ML accelerators don't. As I said above, both have the new CPU, so both should have the SME core. I would be shocked if the A18 didn't have the latest matrix accelerator.
You didn't forget? You mentioned the new media engine and USB 3 and yes they explicitly mention that those capabilities were unique to the Pro.
Now the ISP and a couple of the other points are mentioned separately and not necessarily unique wrt to the A18, but might be. Those were less clear.
I'm far less certain about that, but I lean that direction. Again, 3rd party die shots should confirm.
Last edited:
Yeah, I agree. We can’t know for sure yet. I edited my post to reflect that but you replied in the meantime 😅
Oh I know how that goes! I'm constantly updating what I write and am always checking to see if anyone has replied or even reacted to my post while I'm editing. I still edit even if they have.
