Xpeng’s self-developed autonomous driving chip, named Turing, is ready to enter mass manufacturing within the second quarter and can initially be featured in one of many firm’s new electrical automobile (EV) fashions, as reported by native media.
This chip reportedly outperforms general-purpose automotive chips by 20% in utilization. It possesses the aptitude to handle intensive fashions with as much as 30 billion parameters, a big enhance in parameter capability in comparison with rivals. For context, Li Auto’s Visible-Language-Motion (VLA) parameter depend is roughly 2.2 billion.
Whereas the Turing chip boasts spectacular numbers, bigger mannequin parameters typically deliver challenges, notably concerning latency throughout inference, a priority that Xpeng has but to deal with publicly.
On August 28, 2024, throughout the unveiling of its new mannequin, the Mona M03, Xpeng’s chairman and CEO, He Xiaopeng, introduced the profitable tape-out of the Turing chip on August 23. Tape-out is a time period in built-in circuit design that signifies the transition of a circuit design right into a manufacturable chip after the design course of is accomplished. Following a profitable tape-out, the chip prototype could be mass-produced, contingent on passing additional testing.
He Xiaopeng talked about that Turing chips might be integrated into Xpeng’s EVs, robots, and flying autos. The chip contains a central computing structure that helps each clever driving techniques and good cockpits, with AI computing capabilities comparable to 3 Nvidia Orin X chips. The Turing chip’s computing energy is reported to be round 700 TOPS, inserting it close to Nvidia’s newest Drive Thor chip.
Nvidia’s Drive Thor System-on-Chip (SoC) for autonomous driving was launched on September 20, 2022, as a successor to the Drive Orin, concentrating on automotive fashions for 2025. Initially designed for two,000 TOPS of computing energy, Thor’s mass manufacturing has confronted difficulties, main Nvidia to supply a model with roughly 750 TOPS as a substitute, amid ongoing {hardware} and software program points.
Automakers that originally deliberate to undertake Thor chips, together with BYD, Zeekr, Li Auto, and Xpeng, could also be impacted by these challenges. In-house chip improvement is seen as a means for automotive producers to speed up the introduction of recent fashions.
Nio, one other participant within the EV market, has commenced mass manufacturing of its personal Shenji NX9031 autonomous driving chip, primarily based on a 5 nm course of, with the ET9 being the primary mannequin to make the most of this know-how. Though Nio hasn’t specified the computing energy of the Shenji NX9031, it is claimed to be equal to 4 mainstream good driving chips, with two getting used within the ET9.
In a February reside broadcast, Nio’s founder, William Li, introduced that the corporate’s common fashions, such because the ET5, ET5 Touring, ES6, and EC6, would endure facelifts that incorporate Nio’s Shenji chip. Just lately, rumors indicated that Nio’s up to date ES6 SUV and EC6 coupe are anticipated to launch in Could, adopted by the up to date ET5 sedan and ET5 Touring capturing brake in June.
At the moment, Nio’s principal fashions come geared up with 4 Nvidia Orin chips, yielding a complete computing energy of 1,016 TOPS. The primary mannequin from its sub-brand Onvo, the L60, makes use of one Nvidia Orin X chip with 254 TOPS.
Experiences counsel that Li Auto’s in-house autonomous driving chip is nearing tape-out, and if profitable, the corporate might prioritize its personal developed chip in future autos.
Source link
