英伟达想用AI接管游戏画面，还要让NPC“活”起来

Core Insights - Nvidia is reshaping the gaming rendering process and interaction experience through continuous AI technology iterations without launching new graphics cards [1] - The focus of Nvidia's updates at CES this year was on software, particularly the release of DLSS 4.5 and advancements in NPC technology [1] Group 1: DLSS 4.5 Technology - The key highlight of the release is the update to DLSS (Deep Learning Super Sampling) technology, introducing "Dynamic Multi Frame Generation" and a new 6x multi-frame generation mode [2] - DLSS 4.5 utilizes a second-generation Transformer model to generate up to five additional frames for every traditional frame rendered, significantly increasing the pixel output compared to traditional rasterization or ray tracing [2] - DLSS 4.5 is expected to launch in spring 2024 and will be compatible with all RTX series graphics cards, with optimal performance on the latest RTX 40 and 50 series [4] - When combined with the GeForce RTX 50 series, DLSS 4.5 can achieve over 240 frames per second in 4K gaming with full path tracing enabled [4] - Over 250 games and applications currently support DLSS 4 technology, including new titles like "007: The Beginning" and "Resident Evil: The Resurrection" [4] Group 2: NPC Technology Advancements - Nvidia is enhancing game interaction logic through generative AI, expanding the ACE (Avatar Cloud Engine) technology suite to create autonomous NPCs with perception, planning, and action capabilities [5] - In collaboration with KRAFTON, Nvidia demonstrated an AI teammate in "PUBG" that can remember past player interactions and provide contextually relevant dialogue, with plans for limited testing in mid-2024 [5] - The AI advisor in the strategy game "Total War: Pharaoh" showcases enhanced logical capabilities, offering real-time tactical guidance based on the game's current situation [5] - Nvidia's architecture for NPC memory does not solely rely on cloud models but uses a local vector database to store player behavior and dialogue history, significantly reducing cloud inference costs and addressing privacy concerns [6] Group 3: Cloud Gaming and AI Acceleration - Nvidia's cloud gaming service GeForce NOW announced performance equivalent to RTX 5080 graphics cards and added native support for Linux systems and Amazon Fire TV Stick [8] - For content creators, Nvidia introduced AI acceleration solutions based on RTX GPUs, achieving up to a 3x performance improvement in video and image generation while reducing memory usage by 60% [8] - Nvidia also enhanced the inference speed of small language models like Llama.cpp by 35% and specifically optimized the domestic open-source model DeepSeek-R1 for high inference throughput on RTX graphics cards [8]