The first Neural Engine. It has 2 cores and can perform up to 600 billion operations per second. However, this version of the Neural Engine cannot be used by Core ML. It's only used for tasks such as Face ID and Animoji.
Devices:
- iPhone 8 (2017)
- iPhone 8 Plus (2017)
- iPhone X (2017)
The second generation of the Neural Engine, but the first version that can be used by Core ML (on iOS 12 and up).
This ANE has 8 cores and can perform up to 5 trillion operations per second. Apple claims that Core ML on the A12 is 9 times faster at 1/10th the energy usage compared to the A11.
Devices:
- iPhone XS (2018)
- iPhone XS Max (2018)
- iPhone XR (2018)
- iPad (8th gen, 2020)
- iPad Air (3rd gen, 2019)
- iPad Mini (5th gen, 2019)
- Apple TV 4K (2nd gen, 2021)
This has the same Neural Engine as the A12.
Devices:
- iPad Pro 11-inch (1st gen, 2018)
- iPad Pro 12.9-inch (3rd gen, 2018)
This has the same Neural Engine as the A12X and A12.
Devices:
- iPad Pro 11-inch (2nd gen, 2020)
- iPad Pro 12.9-inch (4th gen, 2020)
- Developer Transition Kit (2020)
This Neural Engine has 8 cores and is 20% faster and consumes 15% lower power than the A12.
Devices:
- iPhone 11 (2019)
- iPhone 11 Pro (2019)
- iPhone 11 Pro Max (2019)
- iPhone SE (2nd gen, 2020)
- iPad (9th gen, 2021)
- Studio Display (2022)
The Performance Core in the A13's CPU also has its own machine learning accelerators (AMX blocks) that do matrix multiplications up to 6x faster than the A12's CPU.
The A14 has a 16-core Neural Engine that is twice as fast as the previous generation, and can perform 11 trillion operations per second.
Devices:
- iPad Air (4th gen, 2020)
- iPhone 12 (2020)
- iPhone 12 Mini (2020)
- iPhone 12 Pro (2020)
- iPhone 12 Pro Max (2020)
- iPad (10th gen, 2022)
The A14 also has second-generation "AMX blocks" for accelerating machine learning operations (matrix multiplications) on the CPU, including Efficiency Cores.
The Neural Engine in the M1 has 16 cores and can perform up to 11 trillion operations per second. It is the first Neural Engine available on a macOS device.
Most likely this is the same Neural Engine as in the A14 Bionic.
Devices:
- MacBook Air (2020)
- MacBook Pro 13-inch with two Thunderbolt 3 ports (2020)
- Mac mini (2020)
- iMac 24-inch (2021)
- iPad Pro 11-inch (3rd gen, 2021)
- iPad Pro 12.9-inch (5th gen, 2021)
- iPad Air (5th gen, 2022)
The Neural Engine is not available on Intel-based Macs, only on Macs with Apple Silicon.
The Neural Engine in the M1 Pro has 16 cores and can perform up to 11 trillion operations per second. It is most likely identical to the M1's Neural Engine.
Devices:
- MacBook Pro 14-inch (2021)
- MacBook Pro 16-inch (2021)
The Neural Engine in M1 Max has 16 cores and can perform up to 11 trillion operations per second. It is most likely identical to the M1's Neural Engine.
Devices:
- MacBook Pro 14-inch (2021)
- MacBook Pro 16-inch (2021)
- Mac Studio (2022)
The Neural Engine in M1 Ultra has 32 cores and can perform up to 22 trillion operations per second.
Devices:
- Mac Studio (2022)
The A15 has a 16-core Neural Engine, with the same amount of cores it can perform 15.8 trillion operations per second (43% faster than the previous generation).
Devices:
- iPad Mini (6th gen, 2021)
- iPhone 13 (2021)
- iPhone 13 Mini (2021)
- iPhone 13 Pro (2021)
- iPhone 13 Pro Max (2021)
- iPhone 14 (2022)
- iPhone 14 Plus (2022)
- iPhone SE (3rd gen, 2022)
- Apple TV 4K (3rd gen, 2022)
The M2 appears to use the 16-core Neural Engine from the A15 Bionic, performing up to 15.8 trillion operations per second.
Devices:
- MacBook Air 13-inch (2022)
- MacBook Pro 13-inch with two Thunderbolt 3 ports (2022)
- iPad Pro 11-inch (4th gen, 2022)
- iPad Pro 12.9-inch (6th gen, 2022)
- Mac mini (2023)
- MacBook Air 15-inch (2023)
- Apple Vision Pro (2024)
The M2 Pro uses the same 16-core Neural Engine from the M2, performing up to 15.8 trillion operations per second.
Devices:
- MacBook Pro 14-inch (2023)
- MacBook Pro 16-inch (2023)
- Mac mini (2023)
- iPad Air 11-inch (2024)
- iPad Air 13-inch (2024)
The M2 Max uses the same 16-core Neural Engine from the M2, performing up to 15.8 trillion operations per second.
Devices:
- MacBook Pro 14-inch (early 2023)
- MacBook Pro 16-inch (early 2023)
The M2 Ultra has a 32-core Neural Engine, performing up to 31.6 trillion operations per second.
Devices:
- Mac Studio (2023)
- Mac Pro (2023)
The A16 has a 16-core Neural Engine, the same amount of cores as its predecessor. However, it can perform up to 17 trillion operations per second (7% faster than the previous generation).
Devices:
- iPhone 14 Pro (2022)
- iPhone 14 Pro Max (2022)
- iPhone 15 (2023)
- iPhone 15 Plus (2023)
The A17 Pro has a 16-core Neural Engine, the same amount of cores as its predecessor. However, it can perform up to 35 trillion operations per second (up to 2x faster than the previous generation).
Devices:
- iPhone 15 Pro (2023)
- iPhone 15 Pro Max (2023)
The M3 appears to use the 16-core Neural Engine from the A15 Bionic, performing up to 15.8 trillion operations per second.
Devices:
- MacBook Pro 14-inch (late 2023)
- iMac 24-inch (2023)
- MacBook Air 13-inch (2024)
- MacBook Air 15-inch (2024)
The M3 Pro uses the same 16-core Neural Engine from the M3, performing up to 15.8 trillion operations per second.
Devices:
- MacBook Pro 14-inch (late 2023)
- MacBook Pro 16-inch (late 2023)
The M3 Max uses the same 16-core Neural Engine from the M3, performing up to 15.8 trillion operations per second.
Devices:
- MacBook Pro 14-inch (late 2023)
- MacBook Pro 16-inch (late 2023)
The M4 uses a 16-core Neural Engine, performing up to 38 trillion operations per second.
Devices:
- iPad Pro 11-inch (2024)
- iPad Pro 13-inch (2024)
The A18 has a 16-core Neural Engine and can perform up to 35 trillion operations per second, similar to its predecessor.
Devices:
- iPhone 16 (2024)
- iPhone 16 Plus (2024)
Like the A18, the A18 Pro has a 16-core Neural Engine, the same amount of cores as the A17 Pro.
Devices:
- iPhone 16 Pro (2024)
- iPhone 16 Pro Max (2024)
The S4 SiP has a 2-core Neural Engine, is the first SiP with a Neural Engine on Apple Watch, and is based on the A12 Bionic's Neural Engine. It supports Core ML in watchOS 6 to enable faster processing of on-device inputs.
Devices:
- Apple Watch Series 4 (2018)
The Neural Engine in S5 has a 2-core Neural Engine and is identical to the S4's Neural Engine.
Devices:
- Apple Watch Series 5 (2019)
- Apple Watch SE (1st gen, 2020)
- HomePod mini (2020)
The S6 SiP has a faster 2-core Neural Engine based on the A13 Bionic.
Devices:
- Apple Watch Series 6 (2020)
The S7 SiP has a 2-core Neural Engine and is identical to the S6's Neural Engine.
Devices:
- Apple Watch Series 7 (2021)
- HomePod (2nd gen, 2023)
The S8 SiP has a 2-core Neural Engine and is identical to the S6's Neural Engine.
Devices:
- Apple Watch Series 8 (2022)
- Apple Watch SE (2nd gen, 2022)
- Apple Watch Ultra (1st gen, 2022)
The S9 SiP has a 4-core Neural Engine. According to Apple, the Neural Engine of the S9 processes machine learning tasks up to two times faster.
Devices:
- Apple Watch Series 9 (2023)
- Apple Watch Ultra 2 (2023)
The S10 appears to be identical to the S9, also containing a 4-core Neural Engine.
Devices:
- Apple Watch Series 10 (2024)
It's important to note that not all new devices have a Neural Engine. The A10 and earlier chipsets do not have an ANE, but are still present in certain devices released recently by Apple.
Devices with an A10 Fusion:
- iPhone 7, 7 Plus (2016)
- iPad (6th gen, 2018)
- iPad (7th gen, 2019)
- iPod touch (7th gen, 2019)
Devices with an A10X Fusion:
- iPad Pro 10.5-inch (2017)
- iPad Pro 12.9-inch (2nd gen, 2017)
- Apple TV 4K (1st gen, 2017)
Devices older than 2016 (using A10 Fusion or earlier chipsets) obviously do not have a Neural Engine.
The Apple Watch Series 3 (2017) and earlier, do not have an ANE.
Intel-based Macs do not have an ANE.