What Nvidia has been working on for years and what is now working satisfactorily, should also be able to use Radeon graphics cards in the future: AI upscaling, which makes the game appear more detailed and sharper than the native render resolution would allow. This saves computing power and increases the frame rate.
AMD is developing its process called “Super Resolution” together with Microsoft on the basis of the DirectX API DirectML (Machine Learning). It should ensure stable frame rates, especially when using power-hungry ray tracing graphics effects. However, the function is not yet available for the presentation of the Radeon RX 6000 graphics card series.
Open source and cross-platform
Compared to media like Anandtech explained AMDthat Super Resolution as part of the GPUOpen project should be open source and work across platforms. The chip manufacturer intends to offer developer tools for game studios within its own FidelityFX repertoire.
DirectML uses neural networks to enhance image quality, for example. This works either via compute shaders or via special hardware cores. Microsoft has been experimenting with AI upscaling, also with the help of Nvidia, for at least a year and a half – in a GDC presentation from March 2019 one employee already called the function Super Resolution.
Upscaling with a neural network
For the technology, Nvidia uses a self-trained neural network that is fed with 16K images of all kinds and, based on the results, learns to upscale lower-resolution moving images. Since Deep Learning Super Sampling (DLSS) 2.0, this has worked quite well. Game developers have to build the feature into their titles themselves.
In one Blog post explains Microsoftthat the combined processors of the Xbox Series X and Xbox Series S game consoles already have hardware cores to accelerate DirectML algorithms and thus go beyond the capabilities of AMD’s “Big Navi” GPUs. Sony is apparently cooking its own soup, like Submitted patents for own AI upscaling reveal.