Apple also offers the MacBook Pro 14 with the M1 Max SoC and the faster GPU, but the performance of the M1 Max is limited in combined workloads and it is not much faster than the base model.
https://www.notebookcheck.net/Apple-MacBook-Pro-14-2021-Laptop-Review-The-performance-of-the-M1-Max-is-limited.612296.0.html
I imagine the type of person who pays so dearly for a product with so many deficiencies, The consumption and battery life is much worse than I thought.
It also lacks other laptops for comparison on the consumption side.
As I expected, the M1 arch doesnt scales up well when more cores are added.
still, for a first try, its ok in its base form. Maybe they will address these issues in the M2
Quote from: kek on April 06, 2022, 20:15:23
As I expected, the M1 arch doesnt scales up well when more cores are added.
still, for a first try, its ok in its base form. Maybe they will address these issues in the M2
What are you referring to? The CB results are not bad for 8+2 with no SMT. Don't forget that the vanilla M1 already has 8 cores. That's not much difference when it comes to scaling. To judge scaling better, we would need to know how much do the efficiency cores bring. Doing some elementary math on CB23 (the relationship seems to hold for CB15 as well), it seems that a Firestorm core brings about 1400 points and Icestorm about 550 and the scaling works out practically perfectly. I took the difference between 6+2 and 8+2 (both MBP14) to get a baseline for Firestorm, interpolated it to 4+2, compared to 4+4 M1 (MBP13) to get a baseline for Icestorm, interpolated further to 0+2 and compared the remainder to the baseline Icestorm and it is almost bang on.
For me, the primary disappointment is battery life. MBP13 set a high bar. That seems to have a lot to do with the display but there was significant deterioration compared to the base model which is going to be down to the processor. That is assuming the numbers are representative.
Just in case someone is interested in the numbers, since I already performed the calculations:
8+2 MBP14: 12386
6+2 MBP14: 9581
Difference: 2805 => 1400 per Firestorm
4+2 interpolation: 9581-2805 = 6776
4+4 MBP13: 7811
Difference: 1035 => 500 per Icestorm
0+2 interpolation: 6776-2*2805 = 1166
Expected: 1035
Discrepancy: 131
So, I decided to bump up the Icestorm to 550. To me, 131 is pretty bang on. It's about 1 %. Whatever scaling effects there might be, they are insignificant. It's less than 10 % of a single Firestorm core.
And anyway, scaling out is primarily a software issue. It comes down to how independent are the individual tasks as coordination is what limits you. CB scales great, architecture will make little difference. In a design of a core, it's more about scaling up. Performance/ frequency, frequency/ power, that sort of thing. In hardware, scaling out is more of an efficiency problem.