Arm introduces a technique for upscaling images called Accuracy Super Resolution, abbreviated as Arm ASR, which is developed from AMD FSR 2 that is already open-source. According to Arm, image upscaling techniques can be broadly divided into two main methods: spatial, which is simpler as it uses images from the same frame for calculations, saving computational power but may result in blurry images – a method utilized in FSR 1. The other method is temporal, which is more complex as it uses images from multiple frames for calculations, consuming more processing power but yielding better quality results. When combined with data from game engines, this method improves the overall output, a strategy employed in FSR 2.
Arm opts for the temporal approach and discovers that the existing solution in the market is FSR 2, which is open-source but designed primarily for PCs and consoles. Consequently, Arm further develops FSR 2 to make it suitable for mobile use. Another advantage of leaning towards FSR is the utilization of AI and settings familiar to game developers from FSR.
Arm’s modifications enable ASR (blue graph) to reduce the power of GPU by around 30-40% compared to FSR 2 (red graph).
Additionally, Arm collaborates with MediaTek to test this technique on the Dimensity 9300 chip, resulting in significant battery savings.
Arm announces plans to open-source the ASR technology under the MIT license, similar to FSR, in the near future.
TLDR: Arm introduces ASR, an upscaling technique based on AMD FSR 2, optimizing it for mobile devices and achieving significant GPU power savings, with plans to release it as open-source soon.
Leave a Comment