M5 Pro and Max unveiled

Apologies if this has already been mentioned and I missed it. Do we think there is any advantage in terms of Performance Per Area from having 6 Super + 12 Performance cores vs 12 Performance (now Super) cores and 6 efficiency cores, as we did previously?

No info as far as I am aware. If I were to guess, the new mid-cores are likely smaller than the large cores. The new CPU complex might even be smaller than the old one, which would be a big win economically.

i don't know which conversation, but not mine, because I've been focused on what it means technically for users rather than naming.

Conversation about the naming is a marketing topic. I do find marketing and business topics interesting too. I've known many brilliant engineers and scientists with great ideas and products, but completely unable to explain them to others . This stuff matters too.

Besides, there are multiple conversations being held in this thread. I think we should be capable of enough multitasking to keep track of off of them without creating too much confusion.

What's really interesting about this framing is that it places more weight on rumors than official announcements and engineering. I'd love for you to explain more about what you are thinking!

If you have a concrete question, I'd be happy to try and answer it to the best of my ability. Right now I am not sure what the question is.
 
So the GPU clock is alleged to be similar to the previous generation. Have they increased the ALU count per core to increase performance?
 
So the GPU clock is alleged to be similar to the previous generation. Have they increased the ALU count per core to increase performance?
It would be good to know. I had hoped we might have a new gpu architecture video, like the one released shortly after the M3’s release.

Losing hope in that happening.
 
So the GPU clock is alleged to be similar to the previous generation. Have they increased the ALU count per core to increase performance?

We already know some of the story from A19. Some integer operations are faster, two FP16 operations can be executed per cycle (doubling the resources), more bandwidth, more efficient resource allocation and scheduling. M5 shows some healthy GPU gains over M4, so I assume the same will apply to new models too.
 
Apologies if this has already been mentioned and I missed it. Do we think there is any advantage in terms of Performance Per Area from having 6 Super + 12 Performance cores vs 12 Performance (now Super) cores and 6 efficiency cores, as we did previously?

Aye I also proposed that it might do so above as another rationale for doing this but we won’t know until we get die shots. Hopefully the new core type and the new fusion die is an incentive for die shot makers to release both base M5 and M5 Pro/Max die shots.
 
Conversation about the naming is a marketing topic. I do find marketing and business topics interesting too. I've known many brilliant engineers and scientists with great ideas and products, but completely unable to explain them to others . This stuff matters too.

For me, I also think the marketing topic is a little easier to argue over at the moment. We have a rough idea of what Apple did technically based on what they've said and claimed, but beyond what's already been said about the approach, I'm happy to wait until we have more concrete details.

No info as far as I am aware. If I were to guess, the new mid-cores are likely smaller than the large cores. The new CPU complex might even be smaller than the old one, which would be a big win economically.

I'd almost bet on it. This really seems to be optimizing floorplan/footprint to me. Enough large cores for the couple single threaded things that benefit that will run at a time, and then more space/power efficient cores to get MT uplift over M4. I'd also guess that the mid-cores are effectively treated much the same as efficiency cores by the scheduler. Low QoS goes to the smaller cores by default. High QoS spills over from the large cores as load increases.

Really, the design of a few high-power cores with a bunch of smaller cores makes a lot of sense to me, based on the sort of loads I've seen over the years. You just aren't going to saturate 10 large cores with single-threaded work that gets the most benefit from it. So if you have enough large cores to handle the bursty work you need them for, and then your smaller cores are fast enough to deliver better multi-threaded performance per watt, or square nm, you go do that. And this new middle core I think is what enables Apple to pursue this approach.
 
I'd almost bet on it. This really seems to be optimizing floorplan/footprint to me. Enough large cores for the couple single threaded things that benefit that will run at a time, and then more space/power efficient cores to get MT uplift over M4. I'd also guess that the mid-cores are effectively treated much the same as efficiency cores by the scheduler. Low QoS goes to the smaller cores by default. High QoS spills over from the large cores as load increases.
I feel much the same way. I've owned both a M1 Max and a M4 Max, and have tried to measure load effects on each. (Both in the 16" MBP chassis, so very similar cooling.) With M1, per-core power at Fmax was so low that Apple could keep frequency quite close to peak under all-core loads. But by M4, that had slipped away; there's a substantial all-core frequency penalty.

If this new mid core adds about as much MT throughput per core as a downclocked big core, while using less area and power, that makes all kinds of sense.
 
Not confusing to you, but I guarantee it’s confusing to many people. Even people here who follow it closely and who know a lot about CPUs (I mean, I designed some of the fastest ones in the world for 15 years), think it’s confusing.

So I will stick to terminology that is clear to anyone in the field, and stick with P and E when talking about a particular chip. To have conversations about these things we shouldn’t need to know what Apple has decided to call them this week - I mean, last week the M5’s P core was a P core, so if we are having a thread about it now we all have to start calling it S cores, and, in reading the historical record, figure out whether the post came before or after Mar 3, 2026 to figure out what core is being talked about?
Two different tech authors for notebookcheck got it wrong:



They both think the new P-cores are just rebranded E-cores and the Efficiency core branding has been dropped.
 
Probably no new information, but ArsTechnica has nice comparison tables for the CPUs:

I'm not really a fan of the term "super" core. Do we get "hyper" cores in a few years then?
I can understand that they want to distinguish between high-performance, more efficent performance, and efficency cores, but still...

Also, this architecture looks similar to current Intel CPUs: A handfull performance cores, with a dozen more efficient cores.
But knowing Apple, I'm guessing that both the efficency and the "medium" performance cores are likely better than Intel's efficency cores (are they still directly derived from Atom cores as in the beginning?).

EDIT: The more modular chiplet design should allow for more interesting M5 Ultra.
EDIT2: M5 is apparently a return of the M1&M2 scheme, where the Max has the same number of CPU cores as the Pro but double the GPU cores. For M3&M4 the number of CPU cores was different for Pro and Max.
 
Last edited:
EDIT2: M5 is apparently a return of the M1&M2 scheme, where the Max has the same number of CPU cores as the Pro but double the GPU cores. For M3&M4 the number of CPU cores was different for Pro and Max.
Basically although the M3 Pro was significantly different from the M3 Max, while the M4 Pro was the same as the binned M4 Max. Now the binned M5 Max doesn't reduce the number of CPU cores and neither does the M5 Pro just as the M1/M2 generation did it. So the M4 was only a little different to the M1/M2 way of doing things while the M3 was the real outlier.
 
Last edited:
Fun fact; The Danish version of the Apple Store page, when you click to compare the CPUs, calls M5, M5 Pro and M5 Max a mix of "Super" and "effektivitetskerner" (efficiency cores). The US website distinguishes that M5 is super and E while Pro/Max are super and P on the same page. The Danish page calls it all Super and E, making no distinction between M5 E and M5 Pro "E" cores.

Of course a mistake on the Danish website but could lend credit to this naming scheme being a somewhat recent idea
 
Two different tech authors for notebookcheck got it wrong:



They both think the new P-cores are just rebranded E-cores and the Efficiency core branding has been dropped.
Well, it is notebookcheck.net. Not my idea of the most reliable information source in the world. ;) They didn't really think this one through, if you ask me. If the new "performance" cores were merely M5 generation E cores, I doubt M5 Pro/Max could beat M4 Pro/Max on MT performance.

I'm not really a fan of the term "super" core. Do we get "hyper" cores in a few years then?
I can understand that they want to distinguish between high-performance, more efficent performance, and efficency cores, but still...
Yeah, it's a bit silly. It would've been more accurate and less confusing to call the new type of core something like a "throughput" core, without changing anything else. I assume something like that didn't happen because modern Apple marketing is so devoted to overusing superlatives. My idea's a bit too bland for their current house style.

Also, this architecture looks similar to current Intel CPUs: A handfull performance cores, with a dozen more efficient cores.
But knowing Apple, I'm guessing that both the efficency and the "medium" performance cores are likely better than Intel's efficency cores (are they still directly derived from Atom cores as in the beginning?).
Intel's efficiency cores are indeed borrowed Atom cores, or at least started that way in Alder Lake (their first heterogenous core CPU). However, Intel likes to repurpose brands and use them for radically different products, so don't get too hung up on memories of what the original Atom product was (slow, slow, slow). The Atom core that went into Alder Lake was (iirc) said to be competitive with the old Skylake core. Respectable performance, even if not competitive with Alder Lake's performance cores, and significantly higher performance (and power, and area) than Apple's E cores.

So in a sense, Apple's new approach is similar, but the details are bound to be different. There's long been hints, for example, that P and E cores within each Apple CPU generation are co-designed - meaning the E core is probably a cut-down version of the P core, with lower frequency targets so its physical design can use more efficient library cells. They probably did much the same here, but with different tradeoffs between performance/power/area than before, producing something that's much stronger performance than an E core without going all-out the way P cores do.

All that means Apple's core types, within a single CPU generation, implement exactly the same ISA. Intel ran into trouble with Alder Lake because the Atom core implemented a different subset of the x86 ISA extensions than the performance core. They ended up having to disable a lot of stuff in the P core (notably, support for AVX512) to make ISA features uniform enough for software to deal with. Apple doesn't have that issue, ISA feature support is always (as far as I know) identical across all core types in a chip.
 
I assume something like that didn't happen because modern Apple marketing is so devoted to overusing superlatives.
[...]
don't get too hung up on memories of what the original Atom product was (slow, slow, slow).

Some comments to the ArsTechnica article mentioned Apple using the brand "Superdrive" twice, thus the use of "super" isn't that new.

Off-topic regarding Atom:
I know that the Atoms starting with Baytrail were fairly good (I still work with some of those cores from time to time), but I'm still guessing that Apple's efficency cores might be better.
 
So in a sense, Apple's new approach is similar, but the details are bound to be different. There's long been hints, for example, that P and E cores within each Apple CPU generation are co-designed - meaning the E core is probably a cut-down version of the P core, with lower frequency targets so its physical design can use more efficient library cells. They probably did much the same here, but with different tradeoffs between performance/power/area than before, producing something that's much stronger performance than an E core without going all-out the way P cores do.

All that means Apple's core types, within a single CPU generation, implement exactly the same ISA. Intel ran into trouble with Alder Lake because the Atom core implemented a different subset of the x86 ISA extensions than the performance core. They ended up having to disable a lot of stuff in the P core (notably, support for AVX512) to make ISA features uniform enough for software to deal with. Apple doesn't have that issue, ISA feature support is always (as far as I know) identical across all core types in a chip.

I am also curious how the topology affects the overall results. If Apples new mid cores are indeed 70% of the big core, that would make them quite similar to Intels efficiency cores. But intel uses a large amount of fire-core clusters while Apple ships denser six-core clusters. With generally higher performance than Intel and a considerably larger shared cluster cache, I’d expect Apple to do better on cooperative parallel workloads.
 
Last edited:
Some comments to the ArsTechnica article mentioned Apple using the brand "Superdrive" twice, thus the use of "super" isn't that new.

Off-topic regarding Atom:
I know that the Atoms starting with Baytrail were fairly good (I still work with some of those cores from time to time), but I'm still guessing that Apple's efficency cores might be better.
Apple E cores are better at being extremely small and sipping power, but worse at performance. They're targeted much further towards the efficiency end of the optimization space. (Though it should be noted here that Intel has also added a third core type - but in Intel's case, since they already had big and mid cores, the newcomer is a small core more like Apple E cores. Also, unlike Apple, Intel makes products with all three classes of core in one device.)

The big step function for Atom performance was indeed Baytrail. The orginal "Bonnell" core used in the first few generations of Atom was a simple in-order dual-issue design; the "Silvermont" core that debuted in Baytrail was a clean sheet reboot with wider and out-of-order execution.

In the many generations since Silvermont, they've kept enhancing it by increasing execution width and so on. It's quite grown up now; some of the biggest and most expensive datacenter Xeon products (the ones with 100+ cores) use nothing but Atom cores.
 
If you're wondering about the practical real world effects of what I've described earlier by designing the new P core, and changing the configuration to 6/12 instead of 12/4:

16 inch:

M5 Max gets 22 hours of live streaming and 16 hours of wireless web, while boosting performance like code compilation by 20% vs M4 Max.

M4 Max gets 21 hours of live streaming, and 14 hours of wireless web.

So as I said earlier, this lets them boost performance efficiency and not force the E core to draw even more wattage in future designs. I think it's very cool!
 
If you're wondering about the practical real world effects of what I've described earlier by designing the new P core, and changing the configuration to 6/12 instead of 12/4:

16 inch:

M5 Max gets 22 hours of live streaming and 16 hours of wireless web, while boosting performance like code compilation by 20% vs M4 Max.

M4 Max gets 21 hours of live streaming, and 14 hours of wireless web.

So as I said earlier, this lets them boost performance efficiency and not force the E core to draw even more wattage in future designs. I think it's very cool!
Couldn’t this be due to the new networking chip?
 
May very well be, but RockRock still has a point that the loss of the efficiency cores at least isn't detrimental to power consumption to a point where it matters
Sure, I was just reacting to “this lets them boost performance efficiency” which I took to maybe be an allegation that the CPU is more efficient.
 
Sure, I was just reacting to “this lets them boost performance efficiency” which I took to maybe be an allegation that the CPU is more efficient.
It is more efficient? P cores probably achieve most of what the S cores do but at a lower wattage. N1 is great, but I seriously doubt it the only reason for 2 hours longer battery life specifically on the Max model.

The Max model is what gets the battery boost. So.... that basically confirms it is more efficient

Keep in mind the Pro was 10/4. The new 6/12 set up with S and P cores allows them to boost performance by 30% multi threaded general tasks while maintaining same battery
 
Last edited:
Back
Top