Digital Foundry He conducted various test to try the technology called Microsoft Auto Super Resolutiona new artificial intelligence-based upscaling technique that the Redmond company is experimenting with starting with the new Copilot+ PCs, with promising results so far.
The technology is limited to the hardware in question, i.e. laptops based on ARM and in particular those with Snapdragon processorbut it is likely that it will soon be extended to other models, although perhaps remaining in the ARM sphere and still requiring the presence of an NPU, or a neural processor for managing AI activities.
There is also a list of officially supported games, but this seems to be mostly a first provisional list that will be expanded later, also because the technology applies universally to games on the hardware in question.
A still limited but promising system for gaming on ARM
The principle on which Auto SR is based is substantially the same as NVIDIA DLSS and AMD FSRor the increase in detail starting from an image with a lower native resolution, through the use of artificial intelligence.
However, this is achieved through rather different systems in one case or the other.
DLSS and FSR are systems that work more in depth and may require specific support also on the code front, drawing on a large amount of data including base image, motion vectors, color information and more, while Auto SR acts almost exclusively starting from the base image.
In essence, it uses “a large-scale super-resolution model AI-basedrunning on the Snapdragon Series X Neural Processing Unit (NPU) to reconstruct image details lost when rendering a game at a lower resolution, to boost frame rates.”
The model, in particular, is called a convolutional neural network (CNN), which provides spatial upscaling between two fixed resolutions. It has been trained to add detail and perform anti-aliasing, all of which is optimized specifically for Copilot+ PC hardware powered by the Snapdragon Series X processor, at least for now.
The Auto SR application adds latency to the game equivalent to the time it takes to execute the model, which is currently 12ms, Microsoft says. latency introduced, based on the tests, it doesn’t seem to be that relevant, even applying V-sync to the upscale.
It’s a much more limited technique than DLSS and the like, as it relies exclusively on 720p images and works mainly in post-processing, without referring to previous frames for anti-aliasing like the TAA system, but the practical results seem promising, at least in some of the games tested.
Whether this has any application outside of the Snapdragon-based Copilot+ PCs remains to be seen, as the application is currently quite limited, but it could open up new avenues for gaming on ARM.
#Microsoft #Auto #Super #Resolution #Tested #Digital #Foundry #Promising #Results #Technology
