Qualcomm’s next snapdragon mobile chip comes into focus with on-device AI

Qualcomm is unveiling the next-generation premium Snapdragon mobile platform without a formal press event, instead steadily disclosing the architecture powering upcoming flagship Android phones. The company recently revealed its new Oryon CPU architecture and a revamped Adreno GPU, and today, it is detailing the Hexagon NPU, which will act as the primary AI engine.

This strategy emphasizes performance and memory locality, with a 5GHz CPU and a new FlexCache architecture that dynamically allocates shared cache capacity based on workload. Dedicated AI engines in the Adreno GPU include Matrix Cores for AI model processing and Adreno High-Performance Memory (HPM) for low-latency storage. The Hexagon NPU features an Element Accelerator optimized for transformer workloads, enhanced shared memory, and support for long-context reasoning, multimodal models, concurrent agents, and low-latency action loops.

Qualcomm's Hexagon NPU is designed to handle large language models by activating only the required parameters, reducing power consumption and memory bandwidth requirements.