NVIDIA Frame Generation also works with AMD FSR 2.0 and XeSS
The latest version of NVIDIA's DLSS (Deep Learning Super Sampling), DLSS 3.0, is based on artificial intelligence, but instead of generating pixels, it generates entire frames using the new Optical Multi Frame Generation technology. Then thanks to the Optical Flow Accelerator present in the new Ada Lovelace architecture, which analyzes two game frames in sequence and, calculating the data of the motion vectors of all the elements present in the images (which are not modeled by the motion vectors present in the engine graphic), predicts the movement of the scene greatly reducing the anomalies that occur when the AI renders elements such as particles, reflections, shadows and lighting. The pairs of ultra-high resolution frames generated by the AI, the motion vectors of the Optical Flow Accelerator and those of the game engine are then fed to a convolutional neural network that analyzes them and automatically generates an additional frame for each rendered frame. from the game.
The AI frame generation technology is separate from the actual DLSS upscaling technology, which allows it to be used even at native resolution or with other upscalers. Precisely this characteristic was taken into consideration by colleagues at Igor's Lab, who paired the technology with AMD FSR 2.1 and Intel XeSS in Spider-Man Remastered on a GeForce RTX 4090.
The results, which you can see in detail in the table below, show that the performance differences between DLSS 3.0 and FSR 2.1 with frame generation technology are quite small, which bodes well for the next FSR 3.0. Intel XeSS, on the other hand, is a bit slower, probably also because it is the latest arrival and still needs several optimizations to be able to catch up with its direct rivals. Unfortunately, it is not possible to verify what the actual moving quality is from simple static screenshots.
Photo Credit: Igor's Lab For now, we just have to see how AMD will handle its version of frame generation technology with FSR 3.0 and whether it will be able to run any GPU like its own. upscaling technology, so as to make a more accurate comparison.
