NVIDIA DLSS 4: A Transformative Leap Forward Discussed at CES 2025
During CES 2025, Bryan Catanzaro, the VP of Applied Deep Learning Research at NVIDIA, shared insights about the recently announced NVIDIA Deep Learning Super Sampling (DLSS) 4. The discussion highlighted significant advancements in key areas such as Super Resolution, Ray Reconstruction, and Frame Generation, offering enthusiasts and gamers an exciting glimpse into the future of graphics technology.
Revolutionizing Super Resolution with AI
Catanzaro elaborated on the innovative transition from traditional convolutional neural networks (CNNs) to sophisticated transformer models in both Super Resolution and Ray Reconstruction. These new models boast enhanced intelligence, enabling them to process larger datasets and make superior decisions, ultimately addressing historical issues like shimmering and ghosting that have plagued NVIDIA DLSS in the past. Notably, the updated Super Resolution model offers a remarkable fourfold increase in computational power compared to its predecessor. While Catanzaro refrained from providing specific estimates on additional rendering time, he emphasized that this advancement represents the premier way to leverage the upcoming Blackwell-powered GeForce RTX 50 series graphics cards set to launch shortly.
Frame Generation: An AI-Powered Redesign
The overhaul of Frame Generation marks a significant shift from relying on Optical Flow hardware acceleration to a fully AI-driven solution. Catanzaro explained the rationale behind this transition, stating:
“When we built NVIDIA DLSS 3 Frame Generation, we absolutely needed hardware acceleration to compute Optical Flow. We didn’t have enough Tensor Cores and we didn’t have an Optical Flow algorithm that was good enough.”
He elaborated that using Optical Flow was a pragmatic choice given past limitations, yet this reliance made enhancing performance problematic. The newly introduced AI-centric approach for Frame Generation in DLSS 4 aims to rectify this challenge, providing a more adaptable and efficient method of delivering high-quality graphics.
Optimizing Performance Across Generations
In an enlightening exchange with Digital Foundry’s Alex Battaglia, Catanzaro discussed the potential for adapting the new Frame Generation technology to older hardware like the GeForce RTX 30 series. He stated:
“I think this is primarily a question of optimization and also engineering and then the ultimate user experience.”
Although the immediate focus is on the RTX 50 series, he expressed optimism about the future possibilities for older GPUs, contingent on optimization efforts and enhancing user experience.
Reflection on Previous Generations and Future Prospects
Previously, when Frame Generation was introduced with the RTX 40 series, Catanzaro hinted at the feasibility of porting it to previous models, albeit with potentially limited benefits due to hardware constraints. The removal of the Optical Flow hardware accelerator in DLSS 4 might pave the way for broader compatibility, yet he cautioned that the higher demands for Tensor Core performance still present challenges for older architectures.
Enhancing Gaming Experience
Beyond the technical enhancements, Catanzaro underscored the significance of separating updated flip metering from CPU processes, resulting in a five- to tenfold reduction in frame time variability. This advancement is anticipated to notably improve frame pacing. He also pointed out that the integration of Reflex 2—another AI-based technology—creates a more ‘connected’ gaming experience, which should resonate with gamers who prioritize low latency.
As NVIDIA continues to push boundaries with DLSS technology, the conversations and insights shared at CES 2025 reaffirm the company’s commitment to innovating gaming graphics and user experiences.
Leave a Reply ▼