Arm China Unveils Next-Gen VPU IP 'Emei' to Power AI-Centric Video Workloads

[Technology Launch] Arm China has introduced the VPU IP 'Emei', compatible with multiple operating systems.

Core Insight: The Role of VPUs Is Being Redefined in the AI Era

As video creation and consumption shift from humans to AI, traditional CPU+GPU architectures struggle to meet the demands of massive video processing workloads. At the 'Linglong Vision' event in Shanghai, Arm China launched the 'Linglong' V560/V760 (codenamed 'Emei'), highlighting its critical role in low-latency, high-throughput scenarios.

Key Metrics: Single-Stream Raw Throughput Approaches 360 MB/s

Take autonomous driving as an example: a single video stream at 8-megapixel resolution, 30 fps, and 12-bit RAW generates approximately 360 MB of raw data per second. If a vehicle is equipped with 12 cameras, the total bandwidth approaches 4.3 GB/s, underscoring the necessity for efficient video preprocessing.

Strategic Foundation: VPUs Must Balance Compression Efficiency and AI Collaboration

The new generation of VPUs is no longer just a codec module; it must also achieve ultra-low-latency compression while preserving meaningful information and support multiple OS environments such as Linux and Android, providing high-quality video streams to power AI training and inference.