Nvidia’s commitment to enhancing its Deep Learning Super Sampling (DLSS) technology has been nothing short of extraordinary. While it’s no secret that Nvidia has been a leader in data center GPUs for years, the revelation of a supercomputer dedicated exclusively to DLSS training—operating non-stop for six years has been a game changer in AI-driven gaming solutions.
At the recent RTX Blackwell Editor’s Day during CES 2025, Brian Catanzaro, Nvidia’s VP of applied deep learning research, unveiled new details about DLSS 4, including its transition from convolutional neural networks to a transformer-based model. While the technical leap is impressive, Catanzaro’s comments about Nvidia’s training process truly underscore the depth of their commitment.

For six continuous years, Nvidia’s supercomputer, equipped with thousands of state-of-the-art GPUs, has been dedicated to refining DLSS. This isn’t just an occasional burst of development; it’s a relentless, round-the-clock effort. Catanzaro explained that the supercomputer isn’t just improving DLSS in the abstract—it’s systematically analyzing failures, like ghosting, flickering, or blurriness, in real gaming scenarios. When the model falters, Nvidia’s team identifies the issues, augments the training dataset with these challenging examples, and retrains the model.
The result is a constantly evolving system that learns from its mistakes. Nvidia’s training datasets are perpetually expanding, enriched with examples of both high-quality graphics and complex problems that DLSS must tackle. This iterative process ensures that the model not only improves but also remains robust across a diverse array of games. After retraining, DLSS is tested against hundreds of games to validate its improvements and refine its performance further.

The move to a transformer model in DLSS 4 is another major milestone. Transformers, known for their success in fields like natural language processing, bring greater efficiency and accuracy to image reconstruction. This shift marks a significant step forward, promising sharper visuals, better motion clarity, and reduced artifacts. Early impressions suggest that DLSS 4 is indeed a game-changer, delivering improved performance not just for the upcoming RTX 50-series GPUs but also for a broader range of users.
What stands out most is the sheer dedication of resources. Nvidia’s supercomputer isn’t just crunching numbers—it’s fundamentally reshaping how AI interacts with gaming. It’s a testament to how far Nvidia is willing to go to ensure DLSS remains at the cutting edge, reinforcing its role as a cornerstone of modern gaming graphics.