It has been about two years since NVIDIA introduced its Deep Learning Super Sampling technology. At first, it could not boast of a great image result, and the games looked washed out. But the second version of the technology significantly improved the image quality while maintaining a high frame rate. And in its current form, this technology is very useful for players. Some games, such as Cyberpunk 2077, are nearly impossible to play in 4K with ray tracing enabled unless DLSS is enabled.
During the announcement of the Radeon RX 6000 series of graphics cards, AMD promised to launch its own version of AI-based ultra-high resolution technology. The company did not provide any details on how it will work and when exactly it will be available. AMD only hinted that its solution could be based on Microsoft DirectML technology, but at the same time, the company said that its technology will be launched as part of the FidelityFX technology suite. As such, it is not yet clear if AMD plans to stick with DirectML and push the gaming market towards a unified standard. Or, on the contrary, she intends to create her own implementation, which will require collaboration with game studios. A new slide from AMD suggests that the company may have opted for the second option.
AMD plans to release a major driver update throughout this spring, according to Prohardver. That said, it is argued that this release will provide an update for two resolution-focused technologies: AMD FidelityFX Super Resolution and Radeon Boost.
AMD Radeon Boost technology does not use machine learning algorithms to improve frame rates. Instead, it dynamically changes the resolution during high-speed gameplay. This means that when there is a lot of movement, the resolution will be adjusted and the performance will improve. This technology is a relatively simple way to increase frame rates, but at the same time, it can lead to unstable performance while playing.
The new and improved Radeon Boost technology is rumored to provide higher image quality by dynamically optimizing resolution on fast moving parts of the image only. For comparison: the current implementation affects the entire frame. This, for example, would solve the problem of low-quality weapons in first person shooters when shooting quickly.