He Xiaopeng: Xpeng Motors is expected to achieve true autonomous driving in 2025

IT Home reported on August 2 that this morning, Alibaba Cloud announced through its official Weibo that based on the Alibaba Cloud intelligent computing platform, Xiaopeng Motors has built the country's largest autonomous driving intelligent computing center "Fuyao" in Ulanqab, which is dedicated to autonomous driving model training.

This afternoon, He Xiaopeng , founder and chairman of Xpeng Motors , posted on Weibo: "With the launch of Xpeng's 'Fuyao' intelligent computing center, the computing power for autonomous driving simulation training will be boosted. I believe that in 2025, Xpeng Motors will begin to remove the word "assisted" from "auto-assisted driving" and enter the era of true autonomous driving."

He Xiaopeng believes that any smart car company will have to develop its own intelligence in the future, so it will have very high requirements for computing power. The computing power cost will rise from the current 100 million yuan level to the 1 billion yuan level in the future. The strong coupling between Xiaopeng and Alibaba Cloud in algorithms and computing power is a must.

Xiaopeng Motors has previously released test videos of Xiaopeng City NGP many times, including urban road tests (U-turns, red light recognition, lane changes, large vehicle response and processing, low-speed following judgment, etc.), road tests in extreme weather environments (dark night and heavy rain), etc. He Xiaopeng said that during the testing phase, they will cover all-weather scenarios as much as possible to allow the product to undergo all-round polishing, so as to tend to higher safety, and on July 28, he also said that the final push of City NGP is not far away.

IT Home learned that Xiaopeng Motors' autonomous driving intelligent computing center "Fuyao" has a computing power of up to 600 PFLOPS, which has increased the training speed of Xiaopeng's autonomous driving core model by 170 times, increased the virtualization utilization of GPU resources by 3 times, and reduced end-to-end communication delay to 2 microseconds.