Qualcomm Introduces Adreno Neural Fusion GPU with Dedicated AI Cores for Next Snapdragon Chip

Introduction of Adreno Neural Fusion GPU

Qualcomm has unveiled its next-generation Adreno Neural Fusion GPU, following the detailing of the Oryon CPU for its upcoming flagship Snapdragon chip. This new GPU is designed to integrate advanced AI processing directly into the graphics pipeline.

Dedicated AI Cores and Memory

The Adreno Neural Fusion GPU will be the first chipset to feature dedicated AI GPU cores, known as "Matrix Cores." These cores are situated within each of the three 1.45GHz GPU slices and are designed to run advanced AI models directly inside the graphics pipeline. Additionally, the GPU includes a dedicated 18MB High Performance Memory (HPM) cache, which provides low-latency memory for tile-based rendering, frame buffers, and compute workloads, keeping them local to the graphics subsystem.

Efficiency and Performance Improvements

The integration of Matrix Cores and HPM aims to reduce latency and improve efficiency by keeping AI processing and data close to the silicon. Qualcomm states that this architecture leads to a 40% power improvement when the technology is enabled. The company highlights that processing AI closer to the GPU results in better efficiency and reduced latency for graphics workloads.

Impact on Graphics and Industry Adoption

The new architecture is expected to deliver improved graphics quality, including upscaled images, stable frame rates, and enhanced efficiency through AI-enhanced rendering built directly into the Snapdragon graphics pipeline. Unity and Unreal Engine have already integrated Qualcomm’s Neural Fusion stack, indicating early industry adoption. Further details about the new chip's NPU are anticipated at the Snapdragon Summit scheduled for September 22-24.