Answering my own question; found this in Apple newsroom press release:
With up to 14 cores, including 10 performance cores and four efficiency cores, M4 Pro also provides phenomenal multithreaded performance.
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
With up to 14 cores, including 10 performance cores and four efficiency cores, M4 Pro also provides phenomenal multithreaded performance.
It contains 10 performance and 4 efficiency cores (as seen in Mini press release).
Last edited:
Y'know, I was all set to flame you for laziness. Good thing I checked before I did. In fact, you're right, this doesn't appear to be on their product "tech specs" or ordering page, which is a fairly stupid oversight on their part. (Edit: I think it's in their newsroom piece though.)
It's in the video though. I was really shocked when they announced it. I was expecting 8P+6E, and could have lived with 6P+8E. 10P+4E drastically exceeded my expectations (and everyone else's, as far as I can tell). I guess they had good intel on, um, Intel's plans early enough to know they needed something to counter the giant wave of E cores in Intel's latest chips. (Edit: That's mostly a joke. I don't think they care much what Intel is doing these days.)
Last edited:
My prediction on the rest of the M4 chip family
| Chip | CPU | Geekbench 6 MC | GPU | Geekbench Metal |
| M4 | 4P + 6E | ~14,000 | 10-core | ~54,000 |
| M4 Pro | 10P + 4E | ~23,000 | 20-core | ~110,000 |
| M4 Max | 12P + 4E | ~26,000 | 40-core | ~220,000 |
| M4 Ultra | (12P + 4E) x 2 | ~50,000 | 40-core x 2 | ~400,000 |
We *think* that... if we believe Gurman. We don't know it.
In any case, the projection for GB6MC is ridiculous. That bench, by design, doesn't scale even close to linearly.
The M4 Max will not be a 16-Core CPU as that is what the M3 Max is. It will be a 20-Core CPU with 16-P & 4E.
Also, the M2 Ultra was a 24-core CPU. M4 Ultra will cap that at 30-core CPU with 20P and 10E
If M4 Ultra is a new monolithic die, not 2 x M4 Max, I think it will have more P cores and less E cores than you predicted. Having 10 E cores on desktops without thermal constraints doesn't make much sense.
Very unlikely.
Plausible possibilities for the M4 Max are: 10P, 12P, 15P, and 18P. 20 and 24P are possible but exceedingly unlikely. 4E is most likely, 6E and 8E is vaguely possible. 16P isn't happening unless it's a totally separate die from the Pro, and even then it's unlikely- 12P and 18P are much more likely.
The P count depends on whether the current cluster is actually 5P, or 6P with one fused off, and whether they reuse that exact cluster setup in the Max. It's very likely it's really 5P, leaving 15P most likely for the Max, 20P an outside chance.
The M1 Ultra had the exact same 20-core CPU with 16P and 4E. My expectation is that Apple will match that same count. M4 Max will not exceed 20-Core CPU. An 18-Core binned version will also be made available.
Last edited:
I think in terms of P cores it is more flexible. Uncut M4 Pro has 10 P cores, but M4 only has four. I doubt that P cores are organized into clusters.
I am quite certain that it two clusters. Using a single 10-core cluster would be quite expensive.
Then you know absolutely nothing about this, and probably shouldn't be offering any opinions. All modern chips have cores grouped into clusters.
M1 had a 4P cluster. M2 was the same. M3 went to 6P clusters, and now with M4 we see Apple using both 4P clusters (in the base chip) and 5P clusters (or, low probability, P6 with 1P fused off) in the Pro.
All of this is proven, not supposed. It's quite visible in all the die shots. You can also detect cluster configurations by measuring inter-core latency, or by looking at clocking. You should read up on this; here's a couple of links to get you started:
Andrei's analysis of the M1 Pro/Max - this has a nice die shot of the M1 Pro and Max
Howard Oakley in Apple Silicon - especially check the "cluster" section
You can also google for "annotated die shot"s of the M2, M3, and M4.
You don't even need to do that. Apple is quite open about these things in the developer documentation. You can also query the core configuration using standard APIs. Here is the output of my M3 Max
Code:
hw.perflevel0.physicalcpu: 10
hw.perflevel0.physicalcpu_max: 10
hw.perflevel0.logicalcpu: 10
hw.perflevel0.logicalcpu_max: 10
hw.perflevel0.cpusperl2: 5
hw.perflevel1.physicalcpu: 4
hw.perflevel1.physicalcpu_max: 4
hw.perflevel1.logicalcpu: 4
hw.perflevel1.logicalcpu_max: 4
hw.perflevel1.cpusperl2: 4Here we clearly see 10 P-cores with 5 P-cores per cluster., hence, two clusters. The defining feature of the cluster is that its CPU cores share the L2 cache. On the E-core side we have a single cluster with four cores.
Your expectation is not meaningful because it's not based on anything relevant. Even if you turn out to be right, it's for the wrong reasons.
The history of the M series shows that Apple is continually playing with core and cluster counts. Even within the same generation - in fact, in *this* generation the M4 and M4Pro have different cluster sizes. So at this point, we have almost no info to go on. If you believe Gurman, the M4 Pro and Max will have the same relationship as the M1 Pro and Max: the Pro will be a mask "chop" of the Max. If that's the case, it is certain that any additional cores on the Max will come in the same cluster size as the Pro. Thus, 15P cores, with an outside chance of 20P. There might be no additional E cores, or there might be another cluster of 4. And of course, there's the possibility that (like the M1 Pro/Max) there will be no additional CPU cores at all, just more GPU, media engines, etc.
But again, that's only if you believe Gurman.
I don't think there's enough data to make any sort of good prediction. The nice thing is, we should know in ~20 hours.
Note that (as leman knows, but didn't mention) those are really 6P clusters. But one of the cores is fused off in each cluster.
Not sure if this is the right thread but now that we’ve seen the iMac and the Mac Mini come standard with 16GB RAM when do we think the MBA will be getting that bump or will it just be M4 machines that get it? Thinking about getting a new MBA but don’t want to pay $250 for the 16GB if something is in the works.
Sure all modern processor cores are clustered, for obvious reasons of interconnects, shared cache, etc. But what you were implying is that M4 Max will reuse the cluster in M4 or M4 Pro as a whole, and therefore projecting M4 Max having multiple of 5P or 6P. I am not sure I agree with that. Since Apple was designing M4 Max as a new die, they have complete design freedom in terms of how many P cores.
PS: sorry missed your post in #126. We are in total agreement then.
Last edited:
Looks like M4 Pro has 5P clusters. The full 14-core version has 2 of 5P clusters, and binned 10-core version has one P-core fused off in each cluster.
Are they? Can we be sure? It is possible that M4 Pro is a chop of the Max. I guess we will see tomorrow.
Interesting thought, but just M4, I think. It's their best seller, but I can't see them doing that (giving the base M3 a memory bump).
One question is will the MBA get M4 at all, or will it go directly to M5? There is good reason to think there will be two Macs that get M5 early (June 2025 at the latest, but more likely early May like the M4 iPad Pro this year). Nobody knows which, but with iMac now out of the running, MBA's odds just went way up. It might still be the Mac Studio or even a new Mac Pro, but that would be unusual, to say the least.
Longest wait is until May or June for the M5 MBA with 16 GB standard. Shortest likely wait is until March (a year after the M3 MBA launch) for the M4 MBA, also with 16 GB standard.
Was not the rumor from Gurman about the mid-tier chip, codename Brava, going to be the M4 Pro; but it would have an interconnect, so two Brava chips would make up the laptop variant M4 Max...?
Therefore:
Therefore:
- M4 Max (laptop variant)
- 28-core CPU (20P/8E)
- 40-core GPU
- 32-core Neural Engine
- 546GB/s UMA bandwidth
vs. CPU clock speed, pro vs. normal ( and soon enough vs. Max), do we know allready something ? has there anything been mentioned somewhere ? Or do we have to wait for first benchmarks to come in ? ( i´ve seen nothing mentioned)
i know, we´ve seen MB M4 benchmarks. That was ~4.4GHz
the mini pro has a higher memory bandwith, which indicates it´s a own design.
In case we know nothing yet, how likely or unlikely is it that the mini-pro is clocked slightly higher than the base M4 ?
i know, we´ve seen MB M4 benchmarks. That was ~4.4GHz
the mini pro has a higher memory bandwith, which indicates it´s a own design.
In case we know nothing yet, how likely or unlikely is it that the mini-pro is clocked slightly higher than the base M4 ?