Qualcomm Discloses Next-Gen Adreno GPU Details Featuring Dedicated Matrix Cores for AI
Qualcomm revealed key details of its next-generation Adreno GPU architecture, introducing three dedicated Adreno Matrix Cores designed to accelerate local AI tasks such as neural upscaling and frame generation. This hardware evolution powers Qualcomm's new Adreno Neural Fusion technology, reducing artifacts like flickering and ghosting while lowering power consumption by up to 40%. By embedding dedicated matrix compute hardware into the GPU and securing native integration with Unreal Engine and Unity, Qualcomm is bringing PC-style neural super-resolution (similar to NVIDIA DLSS) directly to mobile chips. This enables mobile games to achieve higher rendering quality and frame rates while keeping thermal and power consumption profiles under control. The next-gen Adreno GPU adopts a tri-slice architecture with clock speeds pushed to 1.45GHz—up from 1.2GHz in the Snapdragon 8 Elite—alongside a 12% boost in overall energy efficiency and retained 18MB Adreno High-Performance Memory. Paired with Oryon CPU cores that break the 5GHz barrier for smartphones, heat dissipation and power control will be critical challenges for device manufacturers.
## BACKGROUND
Neural upscaling and frame generation use AI models to render graphics at a lower base resolution and intelligently generate additional frames, delivering smoother visuals without taxing standard GPU shaders. While desktop graphics cards have long used dedicated matrix processors (such as NVIDIA Tensor Cores) for these math-heavy AI tasks, mobile processors traditionally relied on generalized computing units or software approximations.