Horizon Robotics founder Yu Kai criticizes inflated PPT computing power claims for autonomous driving chips: Nvidia labels at twice dense computing power, while the industry has hyped it to 6-7 times

Horizon Robotics founder Yu Kai, in a Weibo response to the issue of nominal computing power for autonomous driving chips, pointed out that the industry currently has serious inflation in PPT computing power claims. He cited Nvidia's Orin X as an example: its dense computing power is 127 TOPS, but the nominal value is calculated at twice that, 254 TOPS. Some manufacturers even hype nominal computing power to 6-7 times dense computing power, labeling it as 'equivalent computing power.'

The core of this phenomenon lies in the lack of a unified standard for computing power labeling. Nvidia, as an industry benchmark, has its labeling method imitated by most manufacturers, but the subsequent 'equivalent' concept has gradually deviated from actual physical computing power and evolved into a marketing tool. Yu Kai's criticism directly targets the industry's chronic problem: using inflated numbers to attract customers while potentially masking real performance gaps.

For the autonomous driving industry, inflated computing power claims may mislead automakers in chip selection. If automakers plan algorithms based on nominal values, they may encounter insufficient performance during actual deployment, thereby affecting system safety. Yu Kai emphasized that autonomous driving chips are a 'serious matter,' implying that inflated claims may damage technological trust.

This statement also reflects the intense competition in the autonomous driving chip market. As a domestic chip manufacturer, Horizon Robotics needs to benchmark against Nvidia while also facing 'involution' among domestic peers in parameter competition. Although Yu Kai's criticism has its commercial stance, the labeling standardization issue he raised is worth industry reflection and may promote more transparent computing power disclosure mechanisms in the future.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.