CES 2026开幕前夕 芯片巨头密集上新品、提出千倍AI性能提升目标

Group 1: Nvidia's Innovations - Nvidia launched the Rubin platform, consisting of six new chips, focusing on physical AI and large-scale AI factory construction, targeting autonomous driving and robotics [1][2] - The Rubin platform offers 50 Petaflops of NVFP4 inference performance, with a 5x increase in inference performance and a 3.5x increase in training performance compared to the previous Blackwell platform [2] - The cost of generating tokens on the Rubin platform is reduced to about one-tenth of the previous generation, enhancing efficiency [2] Group 2: AMD's AI Performance Goals - AMD's CEO, Dr. Lisa Su, announced the new MI455X GPU, which offers a 10x performance improvement over the MI355X, with a target of 1000x AI performance increase over four years [4] - AMD predicts global computing infrastructure demand will grow from 1 ZettaFLOPS in 2022 to over 100 ZettaFLOPS by 2025, with future capabilities reaching YottaFLOPS levels [4] - The Ryzen AI 400 series processors were introduced, featuring up to 60 TOPS of NPU performance, emphasizing AI as a default capability in PCs [4] Group 3: Intel's New Offerings - Intel introduced the Core Ultra 3 series processors, the first consumer-grade products based on Intel's 18A process, featuring up to 16 CPU cores and 50 TOPS of NPU performance [5] - The flagship model of the Core Ultra 3 series offers a 60% improvement in multi-threaded performance [5] - Initial consumer laptops with the Core Ultra 3 processors will be available for pre-order starting January 6, 2026, with shipments expected in the second quarter of 2026 [5]