The global competition for the underlying computing power of intelligent vehicles has reached a critical turning point. On May 29, BYD officially announced the "Xuanji A3" chip, China's first vehicle-grade 4nm intelligent driving chip that was fully self-developed, designed, and tested throughout the entire supply chain. This achievement not only breaks the public perception that this new energy giant mainly focuses on electrification technology, but also marks the official entry of domestically produced vehicle AI computing chips into the world's top tier.

On core technical parameters, Xuanji A3 directly matches industry giants like NVIDIA Thor. The chip uses a 16-core CPU with a computing power of up to 420K DMIPS, which can easily manage complex multi-modal logic such as intelligent driving, smart cockpit, and body control. The chip also has a memory bandwidth of 273GB/s, and combined with its self-developed bus, it can achieve nanosecond-level low-latency data scheduling. When three chips are combined, its theoretical peak computing power can easily exceed 2100TOPS.

image.png

However, what truly disrupts the industry with Xuanji A3 is not just its absolute computing power, but its differentiated approach in energy efficiency and practical utilization. Unlike the general-purpose GPU architecture used by NVIDIA Orin or Thor, BYD has firmly chosen a dedicated NPU (Neural Network Processor) approach. General-purpose GPUs must retain a lot of redundant logic such as graphics rendering to be compatible with different customer models; Xuanji A3 hardens matrix multiplication and convolution, among other AI operators, directly at the hardware level, and is deeply customized for its own algorithms. This allows its power consumption per unit computing power to be reduced by 20% compared to similar products, and the algorithm resource utilization is doubled, achieving twice the effective performance with fewer transistors.

This specialized architecture brings significant experience leaps in the real physical world. In real-world testing in the old street of Pingshan, Shenzhen, when facing complex situations such as sudden electric vehicles crossing the road and narrow road U-turns, vehicles equipped with Xuanji A3 demonstrated extremely smooth perception and decision-making capabilities. Based on absolute confidence in the fundamental core architecture, BYD even made an unprecedented "guarantee" commitment at the launch event: during urban navigation, if an accident occurs due to the assistance of the driving system, the platform will fully compensate for the economic losses that should be borne by the vehicle, without any limit.