Qualcomm Details Snapdragon 8 Elite Gen 6 Pro AI Features Across NPU, CPU and GPU

Qualcomm has released further details on the AI-processing capabilities of the Snapdragon 8 Elite Gen 6 Pro, covering upgrades to its new Hexagon NPU. The NPU adds an Element Accelerator, a dedicated hardware acceleration unit optimized for specific operations in AI computing, intended to improve the NPU's efficiency when executing related workloads. The new-generation Hexagon NPU's Large Shared Memory capacity is 50% larger than that of the previous generation. According to Qualcomm, this high-speed shared memory can hold data required during AI model operation, reducing the need to frequently access system memory.

In generative AI workloads, models must process large amounts of context data, some of which must be retained continuously during inference. Qualcomm said that if such data is frequently transferred between the NPU and system memory, it increases latency and consumes more power. By enlarging the high-speed shared memory directly available inside the NPU, the design aims to reduce data movement. The KV Cache is among the affected elements: as a user holds continuous conversations with an AI assistant, the context information the model must retain grows, and the KV Cache scales up accordingly. AI agents must complete multiple steps in sequence, including understanding user requirements, formulating an execution plan, invoking different tools, analyzing returned results and proceeding to the next step, with model state and context needing to be preserved throughout.

Beyond the NPU, the CPU and GPU of the Snapdragon 8 Elite Gen 6 Pro have also been upgraded. Qualcomm previously disclosed that the new flagship platform's CPU reaches a maximum frequency of 5GHz. Qualcomm has introduced FlexCache technology for the CPU, which uses on-chip cache more flexibly so the CPU can obtain data more efficiently. On the GPU side, Qualcomm announced an architectural upgrade to the new-generation Adreno GPU and strengthened its ability to execute AI tasks. The GPU adds Adreno Matrix Cores dedicated to matrix computation. Qualcomm also introduced Adreno Neural Fusion technology to strengthen the integration between AI and graphics rendering; the technology can use AI algorithms to improve in-game visuals and manage power consumption.

The AI-related design of the Snapdragon 8 Elite Gen 6 Pro includes: an NPU with larger shared memory and a dedicated accelerator, a CPU that reduces reliance on system memory through FlexCache, and a GPU with added matrix computation capability that participates in graphics processing through AI technology. The information disclosed by Qualcomm shows that the platform's upgrade priorities include improving AI computing efficiency and reducing memory access and power consumption when running AI locally. Adreno Neural Fusion technology is used to strengthen the integration between AI and graphics rendering and involves AI graphics technologies such as super resolution and frame generation.