ARM Processor MLPerf Benchmarking for AI with Integrated NPUs

This project focused on implementing the MLCommons MLPerf benchmark suite on two different ARM processors with integrated NPUs. MLPerf provides multiple test suites based on device types (tiny, edge, mobile, datacenter) to compare AI model speed and power usage across various platforms. Our team successfully ported the MLPerf Tiny suite to an MCU running an RTOS and conducted speed and energy benchmarks for four AI models utilizing the integrated NPU. Additionally, we ported the MLPerf Edge suite to an MPU running Linux and performed speed benchmarks on the integrated NPU. These benchmarks are essential for differentiating chip vendors’ solutions and enabling them to compete effectively in a new and growing market.