Apple offered no Neural Engine throughput figure, quantifying instead a 17 per cent rise in total system memory bandwidth and linking it explicitly to Writing Tools and Image Playground responsiveness. For on-device model decode, bandwidth is often the binding constraint rather than raw accelerator math.
Apple A18 Pro
The genuine change was memory bandwidth, which Apple tied directly to Apple Intelligence latency rather than to Neural Engine throughput.
Silicon from Apple
10 chips on record, newest first.
-
2026
Apple A20 Pro
2nm Dual 16-core Neural Engine (32 cores total)
2x the compute power to run on-device AI models
Relative only -
2026
Apple M6
2nm Dual 16-core Neural Engine
up to 2x the peak compute over previous generations (Neural Engine)
Relative only -
2025
Apple A19 Pro
Not stated by Apple 16-core Neural Engine plus per-GPU-core Neural Accelerators
Up to 3x the peak GPU compute over the previous generation (a GPU figure, not a Neural Engine one)))
Relative only -
2025
Apple A19
Third-generation 3nm Neural Engine plus per-GPU-core Neural Accelerators
No throughput figure published
None published -
2024
Apple A18 Pro
Second-generation 3nm 16-core Neural Engine
17 per cent increase in total system memory bandwidth
Relative only -
2024
Apple A18
Second-generation 3nm 16-core Neural Engine
Runs machine learning models up to 2x faster than A16 Bionic
Relative only -
2023Relative only
-
2022Absolute
-
2021Absolute
-
2020Absolute