- Published on
A20 Pro Chip AI Benchmark Leaked: Apple iPhone 18 Pro NPU Sees Up to 51.53% Boost
- Authors
- Name
- DeepHorus
Apple's new flagship chip A20 Pro's AI performance benchmark appeared on the GeekBench AI platform on September 12. Data shows that the iPhone 18 Pro (model number iPhone19,2) equipped with this chip achieved a maximum increase of 51.53% over the previous-generation iPhone 17 Pro in the neural network processor (NPU) test. This is the first public result of A20 Pro's on-device AI computing power, following the exposure of CPU and GPU benchmark scores.
GeekBench AI is a commonly used on-device machine learning performance testing tool in the industry. Its benchmark platform currently includes two records for the iPhone 18 Pro, corresponding to the processing results of the Core ML framework calling the GPU and NPU/ANE respectively. Core ML is Apple's machine learning runtime, which can dynamically schedule the CPU, GPU, and Neural Engine to execute AI tasks based on performance, memory usage, and power consumption goals.
Specific benchmark data shows that the iPhone 18 Pro's GPU scores are 12,775 (single precision), 24,077 (half precision), and 26,951 (quantized), while the NPU scores reach 7,002, 55,720, and 75,810. The three scores measure on-device inference performance under FP32, FP16, and INT8 precision respectively, with the quantized score being the closest to everyday deployed on-device model scenarios.
Compared with the previous generation products, the A20 Pro's NPU performance has improved significantly. In the single-precision test, the A20 Pro scored 7,002, an increase of 37.0% over the A19 Pro's 5,111 and 53.0% over the A18 Pro's 4,575. In the half-precision test, the A20 Pro scored 55,720, an increase of 51.5% over the A19 Pro and 69.3% over the A18 Pro. In the quantized test, the A20 Pro scored 75,810, an increase of 50.4% over the A19 Pro and 67.2% over the A18 Pro. As a reference, the NPU scores of Apple's M5 chip (11-inch iPad Pro) are 5,411, 41,100, and 57,553 respectively, and the A20 Pro leads in all tests.
In terms of hardware design, the A20 Pro's NPU adopts a dual 16-core architecture, with a total of 32 cores, and adds a new generation neural accelerator, doubling the 8-bit floating-point operation speed. At the same time, the chip's memory bandwidth has been increased by 50% to support more complex on-device models and inference tasks. These improvements provide a stronger computing foundation for on-device AI applications.
For user experience, this NPU upgrade will bring improvements in three aspects: faster on-device large model inference, including local AI assistants, real-time translation, and text-to-image generation; stronger camera AI capabilities, with algorithms such as night mode, portrait, video stabilization, and object tracking running locally in real time; and a more enduring AI experience, as the NPU's energy efficiency is generally better than that of the CPU and GPU, making battery consumption more controllable.
Industry analysts believe that the leap in A20 Pro's NPU performance reflects Apple's accelerated push in the on-device AI field. As the iPhone 18 Pro series is about to open for pre-order, its actual AI application performance is worth watching.
