Qualcomm’s upcoming premium chip has Adreno ‘Neural Fusion’ GPU

Following Qualcomm detailing the core Oryon CPU that would find itself on the next flagship Snapdragon chip, the chipmaker has announced its next-gen Adreno Neural Fusion GPU.

Qualcomm’s next SoC will be the first chipset to feature dedicated AI GPU cores in the graphics pipeline. These so-called “Matrix Cores” sit snugly in each of the three 1.45GHz GPU slices on the new Adreno GPU. Qualcomm says they act as dedicated local hardware to “run advanced AI models directly inside the graphics pipeline.”

The Snapdragon chip’s Adreno GPU will also feature a dedicated 18MB HPM (High Performance Memory) cache. That capacity sounds humble, but Qualcomm says its HPM “provides large, low-latency memory directly to the GPU, allowing tile-based rendering,
frame buffers, and compute workloads to remain local to the graphics subsystem.”

Both components in the new core architecture help make up the entirety of the Adreno Neural Fusion GPU. For the end user, it means a 40% power improvement with the technology enabled. The company detailed in its press release how developing an SoC with AI processing done in closer proximity to the GPU means less latency and better efficiency.

The trick is keeping the work close to the silicon. Adreno Matrix Cores handle AI models right where the graphics happen, and the 18MB of Adreno High Performance Memory holds the heavy data—tiles, frame buffers, compute—on-chip so it isn’t constantly shuffling back and forth to system memory. Less movement means lower latency and better efficiency, which translates into gaming sessions that last longer.

For the time being, Unity and Unreal Engine have already integrated Qualcomm’s Neural Fusion stack.

The end result should be improved graphics quality with “upscaled images, stable frame rates, and improved efficiency through AI-enhanced rendering built directly into the Snapdragon
graphics pipeline,” as Qualcomm puts it.

Snapdragon Summit is scheduled for September 22-24, and more is set to be announced regarding the new chip’s NPU.