Qualcomm is set to unveil its new Snapdragon 8 Gen 6 processors, including two flagship Elite versions, at the Snapdragon Summit on September 22. These chips feature the updated Oryon CPU architecture, capable of reaching clock speeds exceeding 5 GHz, alongside an enhanced graphics subsystem.
The core of the new platform is the upgraded Hexagon NPU, which incorporates a new Element Accelerator component. This module is positioned near scalar, vector, and tensor processing units to accelerate generative AI algorithms, improving the speed and efficiency of AI agents without increasing power consumption.
The new NPU architecture introduces support for Mixture-of-Experts models, allowing the chips to run AI models with up to 30 billion parameters locally. Increased on-chip memory reduces reliance on main system RAM, enabling the NPU to handle larger contexts and more complex, multi-step agent tasks.
Qualcomm reports a 50 percent performance increase during the pre-processing stage and improved decoding throughput for models using INT4 precision. These advancements are specifically designed to make on-device AI agents more responsive and efficient for complex workflows.
