NVIDIA DLSS Transformer Model Takes Over CNN, Improves Image Quality Across All GeForce RTX GPUs

Sarfraz Khan

Introduced alongside DLSS 4, the new NVIDIA Transformer model will be available on all the RTX GPUs, including the RTX 20, 30, 40, and 50 GPUs.

NVIDIA Shifts from CNN to Transformer Model for Drastic Visual Changes In DLSS, Showcases Improvements in Various Instances, Supports All GeForce RTX GPUs

NVIDIA has been using its CNN (Convolutional Neural Networks) to add more pixels to the existing frames, and while it has evolved noticeably in the past years, the technique is supposedly maxed out. NVIDIA has now introduced its DLSS Transformer Model, which replaces CNNs for better visuals alongside DLSS 4.

Related Story MacBook Neo Racked Up More Than 10% Of RTX Spark’s Two-Year Shipment Estimates In Just Over 3 Months, Making It An Impressive Feat

While Multi-Frame Generation is an exclusive DLSS 4 feature for the RTX 50 series GPUs, the transformer model expands to previous generations as well. As announced in the original post as well as by Jacob Freeman, the Transformer Model will be available across all the RTX GPUs, including the RTX 20, RTX 30, RTX 40, and RTX 50 series GPUs.

This means that even though the DLSS Multi-Frame Generation and Frame Generation are exclusive to RTX 50 and 40 series GPUs, the RTX 30 and RTX 20 series will still enjoy the "Enhanced" DLSS Ray Reconstruction and DLSS Super Resolution through the Transformer model. As per NVIDIA, the DLSS Transformer model uses a vision transformer that evaluates all the pixels in a particular frame and understands the importance of individual pixels. This is repeated across multiple frames to generate detailed pixels for enhanced visuals.

Through this method, the Transformer model uses double the parameters used in CNN and achieves better visuals in gameplay, less ghosting, and smoother edges. NVIDIA demonstrated this by showing a couple of scenes in intensive titles. For instance, the Transformer model helps Ray Reconstruction do a better job than it could with the CNN model. In Alan Wake 2, the mesh shows better stability and significantly reduced shimmering with the Transformer model.

The fan blades are now even more detailed since it was difficult for CNN to understand exactly how to solve the poor clarity in such fast-motion instances. With the Transformer model, there is drastically reduced ghosting, and, in another instance, the reduced shimmering around the cables is clear with the Transformer model.

Essentially, DLSS 4 isn't just increasing the performance of all the RTX GPUs but also deploying NVIDIA's advanced Transformer model for better visual stability. It should be noted that apart from increased performance, NVIDIA also talks about latency reduction. This is due to the Reflex technology which is hardwired into DLSS 4.

Reflex 2 is soon going to be launched for RTX 50 series GPUs for even further latency improvement, but will be available on other RTX GPUs as well.

Sarfraz Khan Photo

About the author: Sarfraz Khan is a hardware reporter with a focus on PC components and the builder community. With years of experience writing about PC hardware and laptops, his work has been featured on several reputable technology publications. Sarfraz's hands-on experience is demonstrated through his first-person accounts of using and comparing different hardware configurations, providing practical and relatable insights for everyday users. His technical analysis is respected by peers in the enthusiast community and has been cited by specialized hardware sites such as Germany's Igor's Lab.

Follow Wccftech on Google to get more of our news coverage in your feeds.

Button