Nvidia DLSS Transformer Model Exits Beta

Nvidia’s latest DLSS SDK Update takes its improved Transformer Model out of beta – Expect more widespread adoption

Nvidia has officially released version 310.3.0 of its DLSS SDK, moving its improved Transformer model out of beta. This new AI model delivers higher levels of image quality for DLSS Super Resolution and Reconstruction, acting as key components of Nvidia’s DLSS 4 technology.

With DLSS 4, Nvidia moved from Convolutional Neural Networks (CNNs) to a new Transformer model. This change enables Nvidia’s new DLSS models to leverage twice as many parameters as Nvidia’s previous CNN models. This enables higher levels of image quality, reduced ghosting, and more detail in motion. In other words, Nvidia’s newest AI models boost DLSS’ image quality.

Now that Nvidia’s Transformer model has exited beta, we can expect more game developers to adopt Nvidia’s Transformer model over its older versions of DLSS. This DLSS model has been available in beta for the past six months, following its unveiling alongside Nvidia’s RTX 50 series in January.

With DLSS 4, Nvidia has enhanced image quality using its Transformer model and increased framerates through its Multi-Frame Generation technology. Over time, we expect Nvidia to continue improving DLSS with new profiles and new/numbered DLSS releases.

It’s unknown how long Nvidia plans to use its Transformer model for DLSS. It will likely improve over time as Nvidia continues to enhance its AI upscaling technology. Nvidia’s DLSS 2 CNN-based AI upscaler has been available for over five years. Will Nvidia’s Transformer model last another five years?

You can join the discussion on Nvidia’s DLSS 4 Transformer model exiting beta on the OC3D Forums.