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.
M3 to M4 vs M4 to M5
- Thread starter macrumor2018
- Start date
- Sort by reaction score
Performance wise I think the jump was equal between the two. But architectural jump etc was bigger for M4 -> M5.
Really curious about temperatures with this new approach.
chars1ub0w
macrumors 6502a
So, Apple has S-, P- and E-cores. The M5 P-core got renamed as the S-core (super).Explains a lot about the new cores etc:
Apple debuts M5 Pro and M5 Max to supercharge the most demanding pro workflows
Apple today announced M5 Pro and M5 Max, the world’s most advanced chips for pro laptops, powering the new MacBook Pro.www.apple.com
"The industry-leading super core was first introduced as performance cores in M5, which also adopts the super core name for all M5-based products."
The M5 series has no E-cores. Those went out after the M4.
"It includes six of the highest-performing core design, now called super cores, that are the world’s fastest CPU core. Alongside these cores are 12 all-new performance cores, optimized for power-efficient, multithreaded workloads."
This could be correct, but I believe the base M5 has super cores and E cores, whereas the pro&max has S and P cores.
Edit: confirmed. Base has S and E.
Makes you wish for another chip with 4S, 6E and 20 GPU cores, doesn’t it? It’s GPU that’s lacking in the base models, imo.
chars1ub0w
macrumors 6502a
Doesn't that mean the base M5 was designed earlier? Then a later design update gave Apple the option to replace E with P. But the base was already in production. Let's see if the base M6 has E anymore.
Not necessarily, all these cores have different purposes. The E-cores on the M5 serve to improve efficiency in a thin and light computer design. The new mid-cores in the prosumer chips are specifically optimized for multicore processing.
My guess: Two aims are being targeted by Apple. First, due to thermal and power limitations on N3B (I have and love the M4 Max, but those fans can spin up when fully pushed), Apple decided that 10-12 “super/performance” cores was no longer appropriate for a laptop. They are hitting the limits here from the M1 design.
Further, if you modeled 6s x 12mid vs 12s x 4e (M4 Max), the lightly threaded things will be unaffected from a core count perspective, and sophisticated multithreaded applications will often favor the higher core count. There will likely be moderately threaded apps that might have benefited more from a higher super/perf core count; but after studying innumerable benchmarks and tests over the years, I’d posit that this is a minority but still meaningful case for modern workloads.
Using super rough numbers, you can even do a little math. Reportedly the m-cores are 70% of the supers. So your weighted average core count is 6 + 12(.7), or 14.4 net cores for M5 Max. We know historically the e cores are more like 1/3 a p-core. So the M4 Max would be 12 + 4(1/3), or 13 1/3 net cores.
Ultimately, even though it won’t always be a winner performance wise, the new core balance will improve battery life and thermal performance overall. Intel’s Panther Lake and Nova follow similar principles. Nevertheless, this oddball new core count is probably why the % increase from M4 to M5 seems a touch higher than M4 Max to M5 Max, using Apple’s very high level estimates. People projected more based on the 12s x 4e continuing.
The second aim, and my bold unoriginal prediction, is that Apple will differentiate the M5 Mac Studio. Expect (1) a higher count of “super” cores, (2) more “m” cores, and (3) at higher configurations, a higher GPU core count. The new fusion / chiplet packaging gives Apple this new flexibility, and authentically, rather than artificially, creates a new desktop product with higher performance vs the current issue of a studio being marginally better than a MacBook Pro except for the Ultra.
Further, if you modeled 6s x 12mid vs 12s x 4e (M4 Max), the lightly threaded things will be unaffected from a core count perspective, and sophisticated multithreaded applications will often favor the higher core count. There will likely be moderately threaded apps that might have benefited more from a higher super/perf core count; but after studying innumerable benchmarks and tests over the years, I’d posit that this is a minority but still meaningful case for modern workloads.
Using super rough numbers, you can even do a little math. Reportedly the m-cores are 70% of the supers. So your weighted average core count is 6 + 12(.7), or 14.4 net cores for M5 Max. We know historically the e cores are more like 1/3 a p-core. So the M4 Max would be 12 + 4(1/3), or 13 1/3 net cores.
Ultimately, even though it won’t always be a winner performance wise, the new core balance will improve battery life and thermal performance overall. Intel’s Panther Lake and Nova follow similar principles. Nevertheless, this oddball new core count is probably why the % increase from M4 to M5 seems a touch higher than M4 Max to M5 Max, using Apple’s very high level estimates. People projected more based on the 12s x 4e continuing.
The second aim, and my bold unoriginal prediction, is that Apple will differentiate the M5 Mac Studio. Expect (1) a higher count of “super” cores, (2) more “m” cores, and (3) at higher configurations, a higher GPU core count. The new fusion / chiplet packaging gives Apple this new flexibility, and authentically, rather than artificially, creates a new desktop product with higher performance vs the current issue of a studio being marginally better than a MacBook Pro except for the Ultra.
Last edited:
was it you that commented somewhere about going from an m4 mini to a studio for gaming?
just ordered a refurb 16/40. pretty excited
going to give my m4 pro mbp to my partner. never really wanted a laptop anyway, beautiful of a machine as it is.
maflynn
macrumors Broadwell
Yes, I had found gaming with crossover a bit disappointing with the Mini, and I moved up to the Studio. I've been very pleased with the Studio, in fact as I've said many times, its been the best desktop computer I've owned
Blue Quark
macrumors regular
Alameda
macrumors 68000
Upgrading from one M-chip to the next year’s chip has never been very sensible, as each year’s performance gains are somewhat modest. But over a period of years, the improvements add up.
The jury is still out on the M5. Apple claims that its AI engine is greatly improved but we won’t know until the parts are in the hands of more people.
chars1ub0w
macrumors 6502a
Well, I am surprised. On my own graph problem code, I thought my 16" MacBook Pro M1 Max would beat my MacBook Pro M3 (base version, not Pro). The base M3 either beats my M1 Max or equals it! And I never use the base M3 machine for computation, thinking it was inferior.
* Single-threaded version of the code, of course, the M3 wins (dotted lines).
* Multi-threaded (6 threads), I thought the M1 Max should win as it can use P-cores only, and the M3 has to use its E-cores on the task. But they are equal!
* Multi-threaded (5 threads). M3 wins.
The graph problem nicely divides into 6 or 30 threads (with some mutex'ing when accessing the graph). M3 Ultra can have 32 CPU-cores, and I'd love to have access to one for a few days.
* Single-threaded version of the code, of course, the M3 wins (dotted lines).
* Multi-threaded (6 threads), I thought the M1 Max should win as it can use P-cores only, and the M3 has to use its E-cores on the task. But they are equal!
* Multi-threaded (5 threads). M3 wins.
The graph problem nicely divides into 6 or 30 threads (with some mutex'ing when accessing the graph). M3 Ultra can have 32 CPU-cores, and I'd love to have access to one for a few days.
Last edited:
From main page:
M5 Max with an 18-core CPU achieved a score of 29,233 for multi-core CPU performance, which tops the 27,726 score achieved by the Mac Studio's M3 Ultra chip with a 32-core CPU. M5 Max is now the fastest Apple silicon chip ever, and it even topped every other consumer PC processor in the Geekbench database.
In terms of multi-core CPU performance, the M5 Max is up to 5% faster than the M3 Ultra, and up to 14% to 15% faster than the M4 Max chip with a 16-core CPU.
Here is a comparison of the multi-core CPU results:
M5 Max with an 18-core CPU achieved a score of 29,233 for multi-core CPU performance, which tops the 27,726 score achieved by the Mac Studio's M3 Ultra chip with a 32-core CPU. M5 Max is now the fastest Apple silicon chip ever, and it even topped every other consumer PC processor in the Geekbench database.
In terms of multi-core CPU performance, the M5 Max is up to 5% faster than the M3 Ultra, and up to 14% to 15% faster than the M4 Max chip with a 16-core CPU.
Here is a comparison of the multi-core CPU results:
- 16-inch MacBook Pro with M5 Max (18-core CPU): 29,233 (one result)
- Mac Studio with M3 Ultra (32-core CPU): 27,726 (average of all results)
- Mac Studio with M4 Max (16-core CPU): 26,166 (average of all results)
- 16-inch MacBook Pro with M4 Max (16-core CPU): 25,702 (average of all results)
chars1ub0w
macrumors 6502a
If I understand you, we should be able to roughly model system performance in this fashion (per core) using Geekbench numbers as a convenient point base.
E.g. if we have 6 threads to run and a mix 4 M3 P-cores and 2 E-cores, we obtain 16767 pts (66%), but under the M1 Max (6 P-cores), we get 14310 pts (56%), and under the M5 Pro (6 S-cores), we get 25358 pts (100%).
If we have 5 threads to run, M1 Max: 11925 pts (56%), M3: 14594 pts (69%), M5 Pro: 21140 pts (100%)
Then that could be used to check (M1 Max vs. M3) and predict (M5 Pro) run-times on lightly-threaded graph problems like mine.
Last edited:
That was a marketing move. The new 's-cores' are the old 'p-cores'. The new kid on the block are the 'm-cores' which apparently Apple was hell-bent on not calling that. Ergo the entire confusion.
Based on the available information, these new cores are the next generation of E-cores, slightly wider and clocked higher. Nevertheless, the improvement in compute throughput is real.
throAU
macrumors G4
(all referring to Pro/max below)
M3 to M4 was significant - M3, especially pro was a little bit of a disappointment from M2. It was better but not better in the way that M2 was vs. M1 generation.
M4 brought the full memory bandwidth back.
M4 to M5 from what i’ve seen is incremental improvement - which is totally fine - with ZERO regressions vs. prior gen.
M2 to M3 generation had some regressions.
throAU
macrumors G4
As someone with both M4 and M5 generation hardware - M5 performance cores are noticeably faster.
Agree with you regarding Pro/Max, although I'd like to see some more tests and real-world usage for M5 Pro/Max before drawing a definitive conclusion.
When talking about the base Mx series, I think M5 has an incredible value proposition. It's not a large jump in performance in paper, but a substantial improvement in actual usability for most users — for example reviews highlight the gaming experience on the M5 MacBook Air.
throAU
macrumors G4
Yeah to be clear, i haven’t compared M4 Pro/max vs M5 Pro/max as i don’t own it (only M4 max).
But holy crap the M5 in the ipad pro smokes. Like… noticeably more responsive than my M4 max in light use.
It’s super responsive. Maybe there’s some ipadOS vs. macOS in there, but honestly… I would bet it’s more to do with the M5 architecture just being that more responsive vs. the M4 generation (based on prior experience with various A/M series hardware below).
Same sort of thing i noticed between the M1 ipad Air i had and the iphone 13 mini. A14 cores (M1) vs. A15 cores in the iphone 13.
Both on iOS - the iphone felt snappier despite being A series vs. M - because the generational improvement in the cores.
Ditto for A15 based ipad mini vs. ipad Air M1. The mini felt snappier under light load.
I think a lot of people have only had experience with maybe 1 generation of M series vs. older A series.
I’ll say it here again (in line with what i’ve said about the macbook Neo) - 2-3 generation leap of Apple Silicon and you’re looking at the lower tier of hardware being faster than 1-2 tiers up overall, and even a single generational jump is significantly noticeable in single thread/light load scenarios. No need to benchmark it, you can instantly feel it in terms of response.
E.g.,. M4 Pro being faster than M1 or M2 Max, etc.
Same holds true for A series vs. base M series. The new A series parts are FAST and make M1 look slow.
Last edited:
It is marketing, obviously. At the same time, it seems like M5 main cores got a crazy IPC improvement over the previous generations. Looking at the Clang GB6 subtest, M5 single-core is 10% faster than M4, but the clock frequency is only 2% higher — that's 8-9% higher IPC on top of an already industry-leading IPC!
Unless there are additional shenanigans we don't see (e.g. that these cores can actually clock higher), from the CPU arch perspective what we see here is completely unprecedented and frankly, entirely insane. Next-best x86 CPU has half the IPC for this particular subtest. Snapdragon/Nuvia Oryon cores are also around 20% slower iso-clock.
Exactly. Apple have been pushing the e-cores' performance more and more until they're not justified to be called e-cores any longer. But "m-cores" would've been fine, without the unneeded confusion of the s-cores.
Marketing is tricky. To a general customer, "mid-core" does not sound as exciting as "performance core". From the revenue optimization standpoint, the naming makes sense (Qualcomm uses "prime" vs. "performance" cores, for example). And regarding confusion — it might be confusing for enthusiasts like ourselves who like to do generational comparisons, but for the most customers the messaging is quite clear: "M5 has super-cores now, so it's even better!". One can of course argue that this labeling is dishonest and manipulative, but as a company, if an optimized messaging helps you to drive more purchase decisions, you'd be stupid not to do it. It's just not a smart stance to take as a business.