Qualcomm’s Adreno Neural Fusion Brings Neural Graphics to Android
Leading up to Qualcomm’s Snapdragon Summit happening later this month in Hawaii. Qualcomm has already disclosed its dual-flagship strategy and its new 5 GHz Oryon mobile CPU. This week, Qualcomm is disclosing some limited details about its new GPU, the first Android GPU with neural rendering capabilities. Now, that said, neural rendering is quite a hot topic as I test DLSS 5’s neural rendering capabilities on NBA 2K27 on the NVIDIA RTX 5090. That is a very different experience from a mobile form factor, where power is a much greater concern, and neural rendering can help save power and improve performance.
Adreno Neural Fusion Is The Future of Mobile GPUs
At the core of Qualcomm’s new Adreno Neural Fusion GPU are the Adreno Matrix Cores, which are very similar to what Apple did with Apple Silicon’s GPU and what we expect to see from the likes of Arm. Having matrix cores in the GPU makes it a powerful center for AI acceleration, complementing the NPU and often offering more AI compute horsepower at the cost of power. The point of integrating these cores into the GPU is to make it more efficient at AI tasks, benefiting both gaming and AI workloads. While many have spoken extensively about the death of the NPU, I believe that this won’t really impact that balance. If anything, it will meaningfully complement the NPU, improving frame rates and image quality.
HPM Returns in Neural Fusion
Qualcomm introduced HPM in the last generation of Adreno GPUs, giving developers more memory to access while performing GPU tasks. HPM gives developers 18 MB of shared cache, enabling more workloads to stay on the GPU rather than go to memory, which consumes more power and adds latency. For much larger workloads, files will likely have to be stored in RAM, since there’s no discrete GPU or dedicated VRAM.
Neural Fusion Empowers Super Resolution and More Gaming
One of the most beneficial features Adreno Neural Fusion enables for gamers is AI super-resolution. This allows you to render the game at a much lower resolution, enabling higher frame rates and better power consumption. Then you upscale the image using AI super-resolution to restore it to its native resolution, with quality and performance varying by method. I have seen firsthand how well super-resolution works on laptops and PCs, and I’m excited to see how soon game developers will be able to take advantage of it. Qualcomm says that both United and Unreal Engine already have integration for this new GPU and the drivers necessary to make it work well. Qualcomm and its OEM partners still need developer support for Adreno Neural Fusion to make the most of its capabilities and allow the GPU to perform at its maximum efficiency.
