Nvidia推出Vera Rubin AI平台,每单位6个定制芯片,提升了主要技术公司采用的AI计算能力和效率,预计产品将于2026年底推出。 Nvidia launches Vera Rubin AI platform with six custom chips per unit, boosting AI computing power and efficiency, adopted by major tech firms, with products expected late 2026.

Nvidia推出了其Vera Rubin AI平台,每个单位有6个定制芯片,旨在提供5倍于AI计算功率和10倍以上处理AI代用品的效率,系统可扩缩成1000个以上机舱。 Nvidia has launched its Vera Rubin AI platform, featuring six custom chips per unit designed to deliver up to five times the AI computing power and 10 times greater efficiency in processing AI tokens, with systems scalable into pods of over 1,000 units. 该平台,包括Rubin GPU、Vera CPU以及新的网络和储存技术,使得能够更快地培训大型AI模型,并改进长对话的性能。 The platform, including the Rubin GPU, Vera CPU, and new networking and storage technologies, enables faster training of large AI models and improved performance in long conversations. 主要技术公司,包括微软公司、亚马逊公司、谷歌公司和CoreWeave公司正在采用这一方法,预计产品将于2026年底推出。 It is being adopted by major tech firms including Microsoft, Amazon, Google, and CoreWeave, with products expected in late 2026. Nvidia还开发了Alpamayo自行驾驶汽车软件的公开来源,并获得了Groq的AI芯片专长。 Nvidia also open-sourced its Alpamayo self-driving car software and acquired Groq’s AI chip expertise. 尽管竞争日益激烈,但中国对老H200芯片的需求依然强劲,出口许可证尚待发放。 Despite rising competition, demand remains strong for older H200 chips in China, with export licenses pending.