During CES 2025, at the RTX "Blackwell" Editor's Day, NVIDIA unveiled a powerful tool: a dedicated supercomputer that has been enhancing DLSS (Deep Learning Super Sampling) for the past six years. Brian Catanzaro, NVIDIA's VP of applied deep learning research, revealed that the company's latest GPUs have been continuously analyzing and improving the technology that has transformed gaming graphics. Catanzaro explained during his presentation on DLSS 4 that NVIDIA's supercomputer works non-stop to enhance DLSS performance by addressing issues like ghosting, flickering, and blurriness in hundreds of games.
DLSS 4 marks the transition from convolutional neural networks to a transformer model that operates locally on client PCs. The ongoing learning process has been vital in perfecting the technology, with the dedicated supercomputer playing a crucial role in this advancement. The resources dedicated to DLSS development are extensive, requiring not just thousands but tens of thousands of GPUs to create a self-improving DLSS model. NVIDIA's CEO, Jensen Huang, has famously stated that DLSS has the ability to predict the future, a claim that will be put to the test when the Blackwell series is launched.
The use of massive data centers to enhance DLSS is an intriguing approach, with each new generation of NVIDIA GPUs significantly accelerating the process.

