At the launch event, XHEART CEO Li Zonglin introduced the groundbreaking X7 Ultra chip, specifically designed for large-scale AI models. The chip features an ultra-large core NPU architecture, achieving intra-core data interaction and interconnection speeds at the terabit-per-second (TB/s) level. Its multi-tiered cache architecture significantly minimizes inference latency and power consumption. Additionally, the chip incorporates a high-precision, large-model-dedicated operator acceleration engine, which optimizes critical operators of Transformer-based large models through hardware-level enhancements. This engine natively supports FP16 and FP32 high-precision vector calculations, ensuring robust computational accuracy.

Through synergistic innovations in both architecture and toolchains, the X7 Ultra chip achieves a tenfold increase in deployable model parameters, enabling seamless and efficient operation of large-scale AI models within automotive environments. Furthermore, the chip boasts an impressive computational utilization rate of up to 98.7%. When compared to leading industry products, it reduces memory bandwidth consumption by 80% and doubles energy efficiency, setting a new benchmark in automotive chip performance.

Notably, the X7 Ultra chip represents China’s first mass-produced automotive-grade CPU based on the Armv9 architecture. This advancement delivers a 20% performance improvement and a 10% boost in energy efficiency. Its large cluster architecture enhances multi-core collaboration and L3 memory access performance. Additionally, the chip integrates a cutting-edge SVE2 vector processing engine, offering superior DSP performance compared to the previous generation’s NEON technology, further solidifying its position as a leader in automotive AI computing.