XHEART Launches X7 Ultra Automotive AI Chip With Armv9 CPU
XHEART has launched and mass-produced the X7 Ultra, an automotive-grade AI chip featuring 700 TOPS of AI compute power and China's first mass-produced Armv9 automotive CPU. Designed specifically for large language models, the chip incorporates an ultra-large core NPU architecture and a specialized operator acceleration engine. This chip shifts automotive hardware design away from traditional multi-model architectures toward a 'large model-defined chip' approach tailored for end-to-end autonomous driving. By optimizing critical Transformer operators in hardware, it enables 10 times larger model parameters to run efficiently at the vehicle edge. The X7 Ultra delivers a 98.7% NPU compute utilization rate and reduces memory bandwidth consumption by 80% compared to industry flagships. It also includes native support for FP16/FP32 vector calculations, multi-level caching for low-latency inference, and an integrated SVE2 vector processing engine.
## BACKGROUND
Modern intelligent vehicles are increasingly transitioning from traditional modular pipelines to end-to-end large language models that handle perception, planning, and control holistically. Traditional automotive chips often struggle with the massive memory bandwidth and unique operator workloads required by these advanced transformer-based models, necessitating custom silicon architectures.