Nvidia admits it has not prepared LPU for the Chinese market to accelerate AI

This spring, Nvidia launched a new AI accelerator, commonly known as LPU, using the development results of the startup Groq it acquired. This week, company representatives had to deny rumors of preparing an LPU version specifically for the Chinese market.

Image source: NVIDIA

Initially, the provocative publication appears on the pages of the publication informationThe company said that Nvidia will prepare an LPU version for the Chinese market that adapts to U.S. export restrictions by the end of this year, and it is said that multiple customers have already placed orders. Nvidia representatives quickly responded to these rumors, explaining Reutersissued the following statement: “The Information’s report on Nvidia LPUs is incorrect. We do not currently sell LPUs in the Chinese market, and we do not have China-specific LPUs in our long-term plans.”.

Such chips should perform well in inference tasks, and in that sense, its emergence in markets outside China is considered very timely. The Information’s initial publication reported Nvidia’s intention to combine LPU operating software with accelerators available in the Chinese market. The Vera Rubin range of solutions originally intended to be combined with Groq 3 cannot be supplied to China due to US export restrictions. This may require tuning the LPU to work with more mature Nvidia components. Regardless, Nvidia denied the news. At the same time, the company’s leaders have repeatedly expressed concerns about the expansion of accelerators in China after Nvidia was forced to withdraw from multiple segments of the local market.