NPU Models

Google Edge TPU

Vendor: Google

Use Case: AI acceleration for IoT and edge computing

Performance: 4 TOPS (Trillions of Operations Per Second)

Power Consumption: ~2W

Precision: INT8 optimized

Intel Movidius Myriad X

Vendor: Intel

Use Case: Computer vision and AI inference

Performance: 1 TOPS

Power Consumption: 1W

Precision: FP16 and INT8

Apple Neural Engine

Vendor: Apple

Use Case: AI processing for iPhones, iPads, and Macs

Performance: 35 TOPS (A17 Pro)

Power Consumption: Integrated in SoC, optimized for low power

Precision: Mixed precision (INT8, FP16, FP32)

Huawei Ascend 910

Vendor: Huawei

Use Case: AI model training and high-performance computing

Performance: 256 TFLOPS (FP16), 512 TOPS (INT8)

Power Consumption: ~310W

Precision: FP16, INT8

Qualcomm Hexagon

Vendor: Qualcomm

Use Case: AI acceleration in Snapdragon mobile chips

Performance: 45 TOPS (Snapdragon 8 Gen 3)

Power Consumption: Integrated in SoC, low power

Precision: Mixed precision (INT4, INT8, FP16)

Samsung Exynos NPU

Vendor: Samsung

Use Case: AI tasks in Samsung Exynos processors

Performance: 40 TOPS (Exynos 2400)

Power Consumption: Integrated in SoC

Precision: INT8, FP16