英伟达否认年底前推出中国特供 AI 芯片：路线图中无此产品

IT House reported on August 22 that The Information reported yesterday that Nvidia planned to ship a customized AI chip in small quantities to Chinese customers before the end of the year. Nvidia subsequently clearly denied it. A spokesman for Nvidia said: "The Information's reports about Nvidia's LPU are untrue. We currently have no LPU sales in the Chinese market, and there are no LPU products for the Chinese market in the product roadmap." The Information said that the chip is a derivative version of the Groq 3 LPU released by Nvidia at the GTC conference in March this year. Developed using Groq-licensed technology, the chip is said to be designed to work with GPUs to speed up the response times of AI chatbots. Since Nvidia's new generation Vera Rubin platform cannot be exported to China, engineers modified the software to enable the LPU to run with processors available in the Chinese market. The report also said that several Chinese customers have placed orders and the products comply with U.S. export control regulations. Note from IT Home: LPU is Language Processing Unit, which is a special chip optimized for the decoding stage in AI reasoning scenarios. The Nvidia Groq 3 LPU is manufactured using Samsung’s 4nm process and comes with 512MB of on-chip SRAM and no HBM memory. In an interview with CNBC in May this year, Nvidia CEO Jensen Huang said that Nvidia had "to a large extent ceded China's AI chip market to Huawei." Related reading: "NVIDIA CEO Jen-Hsun Huang: Samsung Electronics is responsible for Groq 3 LPU chip wafer foundry" "NVIDIA "acquired" Groq for US$20 billion, and employees cashed out US$5 million per employee" "Completing the AI ​​reasoning puzzle: NVIDIA Huang Jen-Hsun reveals the Groq LPU integration roadmap" "NVIDIA releases Groq