On August 2, based on Ali’s cloud intelligent computing platform, XPeng Automobile built the largest domestic autonomous driving intelligent computing center “Fuyao” in Ulanqab, which is dedicated to autonomous driving model training. According to the report, the “Fuyao” computing power can reach 600PFLOPS, the training speed of the core model of XPeng autonomous driving increased by 170 times.
One PFLOPS (petaFLOPS) is equal to one trillion (=10^15) floating point operations per second. According to the 59th Supercomputer Top500 list announced at the ISC 2022 International Supercomputing Conference in May this year, the second-ranked Japanese “Fuyue” has an arithmetic power of 442 PFLOPS. China’s most powerful supercomputer, Shen Wei – TaihuLight, ranks sixth, with an arithmetic power of 93 PFLOPS.
It is understood that XPeng began to reserve R & D talent in automatic driving early, 2017 from Tesla poached the Autopilot machine learning head Gu Junli, as vice president of assisted driving research and development, but two years left. Later on, the head of the Autopilot R&D team, Wu Xinzhou, was poached from Qualcomm and is still the No. 1 position of XPeng Autopilot.
In the field of autonomous driving, XPeng belongs to the more aggressive of the three in the Azure small Li, such as the earliest switch to Nvidia chips, the earliest on LIDAR, the earliest start of urban NGP (intelligent navigation assisted driving) and so on.
On June 27, XPeng car CEO He Xiaopeng put out a video of his family experiencing the new version of XPeng car city NGP, demonstrating 180-degree turnaround, traffic light recognition, lane change detour, extreme weather/special situation response, unprotected left turn, large vehicle response processing, low-speed following judgment and other scenarios often encountered when driving in the city, you can see that the degree of completion has been very high.