Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

gradi

macrumors 6502
Original poster
Feb 20, 2022
285
156
Are M3 efficiency cores faster than M1 performance cores?

Are M2 efficiency cores faster than M1 performance cores?

Are M3 efficiency cores faster than M2 performance cores?


The M4 is in the new ipad and is expected very soon in Macbooks.

I am just wondering if the efficiency cores may be as fast or faster than a generation or two back performance cores.
 

Malus120

macrumors 6502a
Jun 28, 2002
696
1,455
Are M3 efficiency cores faster than M1 performance cores?
Yes
Are M2 efficiency cores faster than M1 performance cores?
Yes
Are M3 efficiency cores faster than M2 performance cores?
Yes
The M4 is in the new ipad and is expected very soon in Macbooks.

I am just wondering if the efficiency cores may be as fast or faster than a generation or two back performance cores.

While the E-Cores are improving each generation, they have not yet reached the level of performance of an M1 P-Core. This is in part by design, Apple is running them at a really low clock speed to save power. Interestingly they can actually run quite a bit faster (2x faster) as evidenced by the 2 E-Core cluster in the M1 Pro/Max being just as a fast as the 4-Core cluster in the base M1 (due to being allowed to run at double the clock speed)

Anyway, for an idea as to the gen on gen performance improvements, have a look at the Cinebench R23 benchmarks in this review:
If you multiply the Single Core Score by 4 (the number of P-Cores) you'll see that the 4 E-Cores are adding more to the final multi-core score each generation.

You can read more about the P and E cores at

There are some really interesting articles that go into the difference between P and E cores.
 
  • Angry
Reactions: NetMage

gradi

macrumors 6502
Original poster
Feb 20, 2022
285
156
While the E-Cores are improving each generation, they have not yet reached the level of performance of an M1 P-Core.
I am confused about your reply. You answered yes to all 3 of my questions, but then in the sentence above you say the opposite. Did you mean to answer no to all 3 questsions?
This is in part by design, Apple is running them at a really low clock speed to save power. Interestingly they can actually run quite a bit faster (2x faster) as evidenced by the 2 E-Core cluster in the M1 Pro/Max being just as a fast as the 4-Core cluster in the base M1 (due to being allowed to run at double the clock speed)

Anyway, for an idea as to the gen on gen performance improvements, have a look at the Cinebench R23 benchmarks in this review:
If you multiply the Single Core Score by 4 (the number of P-Cores) you'll see that the 4 E-Cores are adding more to the final multi-core score each generation.

You can read more about the P and E cores at

There are some really interesting articles that go into the difference between P and E cores.
Thanks for the links. I will look at them.
 
  • Like
Reactions: NetMage and Xiao_Xi

name99

macrumors 68020
Jun 21, 2004
2,407
2,308
I am confused about your reply. You answered yes to all 3 of my questions, but then in the sentence above you say the opposite. Did you mean to answer no to all 3 questsions?

Thanks for the links. I will look at them.
"Faster" is relative to a workload.
Code that is primarily limited by
- IO
- network
- constant reference to DRAM (ie does not cache well)
- unpredictable branches

Will not run much faster on a P core than an E core. For example a lot of OS code is in this category, and it makes sense to usually run OS code on the E core.

Apple continues to experiment to try to characterize what types of code run best on E cores. Theoretically you can predict what should happen, but you can't be quite sure till you experiment!
Obviously the experiments have shown that there is more of such code than was first guessed (hence first the M3 Pro with 6 E cores, now looks like everyone will get 6 E cores [and maybe 4 Pro gets 8 E cores, to see how well that works?])
And there are obvious cases where P is the winner, basically anything that consists of heavy-duty compute. So compiling, HPC, a JS-heavy web page, stuff like that.

Honestly no-one knows how this will play out. Apple, Intel. AMD, ARM, Qualcomm will all introduce different combinations of P vs E cores, and it's stupid to insist dogmatically that one of these combinations is the *right* one.
If I'm compiling, do I want 6 P and 6 E cores, or 4 P and 12 E cores? Could even depend on the patterns of my code (do I frequently modify .h files that are included a lot, or mostly modify a single .c file?)
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.