Nvidia has released DLSS 4.5, an update accessible via the latest stable version of the Nvidia App. This enhancement allows over 400 games to leverage a new Super Resolution upscaling model, showcasing improved image quality.
New Model and Performance
The new DLSS 4.5 incorporates Model M, which utilizes a second-generation Transformer AI. This marks a progression from the convolutional neural network (CNN) in DLSS 3 and the first Transformer in DLSS 4.0. Nvidia states that Model M requires five times the computing power of its predecessor, having been trained on a significantly larger high-fidelity dataset.
The improvements in image quality are subtle but discernible when examining details like lighting, edges, and motion clarity. The upscaling capabilities from low resolutions, such as 240p to 720p, are noteworthy, though not groundbreaking compared to previous DLSS versions and competitive solutions like AMD FSR and Intel XeSS.
Current Features and Future Developments
DLSS 4.5's current offering includes the advanced Super Resolution Model M. Users need to update their Nvidia App, navigate to Graphics > Global Settings, and choose "DLSS Override - Model Presets" to access the latest enhancements.
Potential performance impacts on older GPUs may arise from increased Tensor-core load requirements, though recent tests on an RTX 4080 Super indicate negligible differences. Looking ahead, Nvidia anticipates releasing new modes and dynamic frame generation.
