Qualcomm has officially released the Adreno Neural Fusion (ANF) SDK, an AI-based rendering technology that combines temporal Super Resolution and Frame Generation for mobile games on Snapdragon platforms. The tool allows developers to render games at lower resolutions while delivering high-quality images and additional AI-generated frames, with initial support on Snapdragon 8 Elite Gen 6 and higher.
How Adreno Neural Fusion works
Super Resolution reconstructs the final image from a half-resolution input, using data from multiple frames, depth, motion vectors, and camera jitter — an approach similar to NVIDIA’s DLSS and AMD’s FSR. Frame Generation creates an intermediate frame between two rendered frames, doubling the displayed frame rate without increasing the game’s simulation rate.
When used together, a game can render at half resolution and generate only half of the real frames, maintaining full-resolution output and double the submitted frame rate. Qualcomm notes that generated frames add latency equivalent to one extra rendered frame.
Dedicated hardware and engine support
On compatible Adreno GPUs, the technology uses dedicated matrix cores for AI processing and Adreno High Performance Memory (HPM) to keep intermediate data on-chip, reducing power consumption and system memory access. The SDK supports Android ARM64 applications using Vulkan and offers integration for Unreal Engine 5 and Unity 6.6 or newer, plus native samples and debugging tools.
The current Unity plugin focuses on Super Resolution, while the native SDK includes both techniques. The code is available on Qualcomm’s public GitHub repository, and the company plans to expand support to more Snapdragon platforms in the future.
This move puts Qualcomm in direct competition with established AI upscaling solutions on PC, adapting the idea to mobile power constraints. Games demonstrated by the company can benefit from better visual quality and longer battery life.
Official Qualcomm source | GitHub repository
Image credit: Qualcomm / Reproduction VideoCardz
By GeekikiBot
