The past year has seen rapid technological advancements, especially in AI, with NVIDIA Research at the forefront. The team, composed of numerous talented individuals, has made significant strides not only in AI but across various technology fields. They have improved GPU performance through breakthroughs in circuits, memory architecture, and sparse arithmetic, advanced real-time rendering, and developed methods to enhance AI efficiency, reducing energy consumption and GPU cycles while improving results. A highlight of the year was generative AI's evolution. NVIDIA can now generate diverse content forms, including 3D models, music, and sounds, with improved control over the output. Generative AI has been applied to science, resulting in highly accurate weather forecasts, precise predictions of blood glucose level responses to foods, and development in autonomous vehicles and robotics. Key initiatives include ConsiStory, which generates multiple images with consistent characters for storytelling, and Edify 3D, facilitating quick 3D object creation for virtual environments. Fugatto, a flexible AI model, can create or modify music and sounds.

GluFormer predicts future blood glucose levels from past data, allowing for precision nutrition. LATTE3D generates 3D shapes from text prompts almost instantaneously. MaskedMimic reconstructs realistic humanoid robot movement. StormCast is a generative AI model that enhances weather prediction accuracy. Throughout 2024, NVIDIA's AI models set records in AI training, inference, route optimization, autonomous driving, and robotics. NVIDIA cuOpt set benchmarks in logistics optimization; the Blackwell platform excelled in AI training and inference benchmarks; Hydra-MDP ranked first in autonomous vehicle driving challenges; and FoundationPose led in estimating and tracking 6D object poses. NVIDIA Research continues to innovate and expects even more groundbreaking results in the coming year.