老黄All in物理AI！最新GPU性能5倍提升，还砸掉了智驾门槛

Core Viewpoint - NVIDIA is shifting its focus entirely towards AI, marking its first appearance at CES in five years without showcasing gaming graphics cards, indicating a strategic pivot towards AI technologies [2][4]. Group 1: New AI Products and Architectures - NVIDIA introduced the next-generation Rubin architecture GPU, which boasts inference and training performance that are 5 times and 3.5 times better than the Blackwell GB200, respectively [4][17]. - The company unveiled five new product families targeting various AI applications, including the NVIDIA Nemotron for Agentic AI, NVIDIA Cosmos for physical AI, and a new model family for autonomous driving called NVIDIA Alpamayo [7][8][29]. - The Vera Rubin NVL72 architecture was officially launched, featuring six core components designed to enhance AI data center capabilities, including the Vera CPU and Rubin GPU [15][17]. Group 2: Performance Metrics and Innovations - The Rubin GPU achieves an inference performance of 50 PFLOPS and a training performance of 35 PFLOPS, significantly enhancing computational capabilities for AI applications [17][25]. - Each Rubin GPU is equipped with 288GB of HBM4 memory and a bandwidth of 22 TB/s, supporting the demands of large-scale AI models [17]. - The NVLink 6 interconnect technology increases the inter-GPU bandwidth to 3.6 TB/s, facilitating efficient communication between expert modules in large models [17]. Group 3: Open Source Initiatives - NVIDIA announced ongoing contributions to open-source training frameworks and multimodal datasets, including 100 trillion language training tokens and 100TB of vehicle sensor data [8][10]. - The Alpamayo model for autonomous driving is the world's first open-source, large-scale visual-language-action inference model, designed to enhance vehicle decision-making capabilities [29][31]. Group 4: Industry Applications and Collaborations - The new models and frameworks are expected to be integrated into various industries, with the Alpamayo model set to debut in the Mercedes-Benz CLA in 2025, showcasing NVIDIA's commitment to advancing autonomous driving technology [29][34]. - The Nemotron models are tailored for specific applications such as speech recognition and safety, enhancing the reliability and efficiency of AI systems [37][39]. - The Cosmos model has been upgraded to generate synthetic data that adheres to real-world physical laws, with applications in robotics and autonomous driving [41][44]. Group 5: Healthcare and Life Sciences - NVIDIA Clara is focused on healthcare and life sciences, aiming to reduce costs and accelerate treatment solutions, with specialized models for protein design and drug discovery [48][49]. - The company is providing datasets of 450,000 synthetic protein structures to researchers, further supporting advancements in personalized medicine [49][50].