理想自研智驾芯片上车路测，部分计算性能超英伟达 Thor-U

Li Auto's self-developed intelligent driving chip M100 has completed the sample return in the first quarter of this year, passing a key stage before mass production. Subsequently, the M100 completed functional and performance testing within two weeks, followed by stress testing conducted by Li Auto's R&D personnel. Currently, the M100 has been used in small batch sample vehicles for road testing.
According to our understanding, the M100 exhibits specific performance characteristics when handling different types of computing tasks: for instance, when running large language model (LLM) computing tasks, the effective computing power provided by one M100 is roughly equivalent to that of two NVIDIA Thor-U chips; whereas, when processing traditional vision tasks related to convolutional neural networks (CNN), such as image recognition, the effective computing power provided by one M100 can be compared to that of three NVIDIA Thor-U chips. (Late Auto)