Qualcomm has outlined a redesigned Hexagon NPU for its upcoming Snapdragon 8 Elite Gen 6 Pro, positioning on-device AI as a central part of its next flagship mobile platform. The new design adds dedicated hardware for more complex AI models, including generative and agentic systems that handle wider context, multiple tools and parallel tasks.

  • A new accelerator optimized for transformer-based AI models, working alongside scalar, vector and matrix extensions.
  • A 50% increase in shared NPU memory, intended to keep frequently used model data closer to the processing hardware.
  • Up to 50% faster INT4 prefill, the stage in which a model processes an initial prompt and its context before beginning to generate a response.
  • Support tailored to Mixture-of-Experts models, which activate only the required portion of a model at each step to reduce resource use.

The teaser arrives as mobile-chip competition increasingly extends beyond conventional CPU and GPU performance. Qualcomm is highlighting the NPU changes as Apple’s A20 Pro has recently appeared in Geekbench with strong results, adding another front to the flagship silicon race.

Diagram illustrating Qualcomm's redesigned Hexagon NPU architecture

Qualcomm is scheduled to unveil its new top-tier Snapdragon platforms on September 22. The announcement should clarify the full Snapdragon 8 Elite Gen 6 Pro specification set, including how the company measures the promised AI gains in shipping devices.

VIAgsmarena.com
Previous articleUGREEN Expands Into Local AI Smart Homes and NAS Storage Up to 272TB
Next articleInsta360 Luna Pro pairs a 1-inch sensor with 8K Dolby Vision capture