Qualcomm is introducing its Snapdragon 8 Elite Gen 6 SoC as well as the Snapdragon 8 Elite Gen 6 Pro at its annual Snapdragon Summit on September 22.

Ahead of that event, today the company is teasing the improvements it's made to one of the chips' Hexagon NPU (presumably we're talking about the Hexagon NPU of the Snapdragon 8 Elite Gen 6 Pro).

This is built for agentic AI, with a transformer-focused Element Accelerator and a 50% larger shared memory that "keeps frequently accessed model data close to the NPU" so that agents "can stay responsive as they juggle longer context, more tools, and concurrent tasks". So you can expect reduced memory bottlenecks and "faster, more responsive agentic AI experiences".

The Element Accelerator is "purpose-built for the transformer workloads that power modern generative and agentic AI", and together with the scalar, vector, and matrix extensions, it accelerates the operations that matter most for large models, helping agents respond faster, as well as reason more efficiently and "deliver richer experiences without compromising mobile power efficiency".

Qualcomm also says the new NPU has up to 50% faster prefill for INT4 models, and it's very well suited to Mixture-of-Experts (MoE) models which only activate a fraction of their parameters per token. The new Hexagon NPU is "designed for always-running AI, long-context reasoning, multimodal models, concurrent agents, and low-latency action loops".

https://www.gsmarena.com/qualcomm_outlines_the_improvements_in_the_snapdragon_8_elite_gen_6_pros_new_npu-news-74569.php