MacBook Pro 2021 review: portable Mac Pros – MacGeneration

“With its M1 chip”, we said last year, “Apple is simply resetting the comparison scales” laptop performance. Since the M1 Pro chip has twice as many graphics cores as the M1 chip, and the M1 Max chip in turn has twice as many graphics cores as the M1 Pro chip, it was obvious that the same sentence could apply. the graphics performance of mobile workstations.

But before measuring the new 14 “and 16” MacBook Pro, we had not yet realized that we would have to change the scale again, and compare these machines to the Mac Pro! “Hang on”, “A new dimension”, “A monster of power”, “The horses are let go”… Choose the metaphor you want: it is true.

Two and two make four

With the M1 platform, Apple is proving that it perfectly masters the principle of “intellectual property blocks”, these circuits that can be stacked like bricks to form a chip. Without offending to Johny Srouji’s teams, we can see the M1 Pro and M1 Max chips as recombinations of the circuits of the M1 chip.

Thus, the processor uses the “thrifty” Icestorm cores and the “high-performance” Firestorm cores, which we are starting to get to know well. The M1 chip combined four cores of each variety, but only used Firestorm cores at the last end, so average performance was mostly determined by the smaller Icestorm cores.

However, if they are surprisingly powerful for supposedly thrifty cores, they are significantly less efficient than the large Firestorm cores. The M1 Pro and M1 Max chips have only two Icestorm cores united in an “E-cluster”, but up to eight Firestorm cores, in the form of a pair of “P-clusters” each benefiting froman L2 cache of 12 MB.

The amount of system cache is tripled on the M1 Pro chip (24MB) and sixfold on the M1 Max chip (48MB). In theory, the expansion and complexity of circuits lengthen data transfer time and increase latency. In practice, the distribution of tasks between the two clusters allows you to enjoy a bandwidth of more than 200 GB / s, a figure worthy of the best server-class chips.

Except that these chips are designed for laptops. The frequency of the cores varies according to their use, from a base frequency of 600 MHz up to 2064 MHz in five steps for Icestorm cores and up to 3228 MHz in fifteen steps for Firestorm cores. Each P-cluster can manage the frequency of its cores independently of the other, the maximum frequency decreasing with the number of active cores.