Apple silicon Neural Engine rankings by generation
Neural Engine leader: the M5 family at 420 (about 4.2× M1). Apple ships one Neural Engine design per generation, so base, Pro, Max, and Ultra chips of the same generation share a single score — this list has one row per generation.
Each bar spans the row; length is relative to the top score on this list. The scale is MacDiff’s own: entry-level M1 Air = 100, not a raw benchmark point.
-
1M5 familyleader420M5 · M5 Pro · M5 Max · 16-core Neural Engine
-
2M4 family329M4 · M4 Pro · M4 Max · 16-core Neural Engine
-
3M3 family226M3 · M3 Pro · M3 Max · M3 Ultra · 16/32-core Neural Engine
-
4M2 family203M2 · M2 Pro · M2 Max · M2 Ultra · 16/32-core Neural Engine
-
5M1 family100M1 · M1 Pro · M1 Max · M1 Ultra · 16/32-core Neural Engine
Why chips of one generation tie
Moving from Pro to Max does not buy a faster Neural Engine; only a newer generation does. Ultra parts fuse two dies with 32 Neural Engine cores, yet MacDiff scores them with their generation, the same as the 16-core chips.
The index covers on-device ML that runs on the Neural Engine (Core ML, Apple Intelligence features). Large models that run on the GPU follow /gpu and memory size instead.
Chip families
Filter the catalog by chip name to see every Mac that ships it.
A18 Pro · M1 · M1 Max · M1 Pro · M1 Ultra · M2 · M2 Max · M2 Pro · M2 Ultra · M3 · M3 Max · M3 Pro · M3 Ultra · M4 · M4 Max · M4 Pro · M5 · M5 Max · M5 Pro · M6
Related
FAQ
Is this Geekbench?
No. These are MacDiff Neural Engine indices (M1 Air = 100). Methodology: /legal/disclaimer.
Is the M4 Max Neural Engine faster than the M4’s?
No. Base, Pro, and Max chips of one generation share the same 16-core Neural Engine, so MacDiff gives them one score. Only a newer generation moves the number.