NVIDIA's new DLSS 5 technology can provide significantly more realistic lighting in games, but it comes at a performance cost. As tests with NBA 2K27 showed, using 3D-Guided Neural Rendering can reduce frame rates by approximately 60% even on GeForce RTX 5080 and RTX 5090 graphics cards at 4K resolution.
One reason for such a high load is the placement of DLSS 5 Neural Rendering in the standard graphics pipeline. At the current stage, the technology works after DLSS Super Resolution. For example, when using Performance mode in 4K, the image is first rendered in 1080p, then scaled to 4K, and Neural Rendering processes the final frame at full resolution.
Meanwhile, NVIDIA continues to optimize DLSS 5 for RTX 50 series graphics cards and has not yet opened official support for the technology for RTX 40 and older generations. However, enthusiasts have already found a way to significantly reduce performance losses.
The modification is called Neural Upstream. It changes the processing order of DLSS 5 Neural Rendering and DLSS Super Resolution: Neural Rendering is performed before image scaling. Thus, at 4K resolution and DLSS Performance mode, the neural network stage works with the original 1080p frame, after which Super Resolution upscales it to 4K.
Neural Upstream is already available as part of the DLSS 5 Autopilot tool, which allows DLSS 5 to be enabled in a large number of PC games.
A significant result was obtained on a GeForce RTX 4080 in Assassin's Creed Shadows. At 4K resolution with DLSS Performance and DLSS 5 Neural Rendering enabled, the standard processing option provided about 31 fps. After activating Neural Upstream, the indicator increased to 50 FfpsS.
The increase was approximately 61%. However, this approach has a significant drawback — a reduction in image quality and detail. In other words, Neural Upstream offers a noticeable performance boost at the cost of some of the benefits for which DLSS 5 was created.