At an October 29 launch event, XHEART (Xinhang Tu) CEO Li Zonglin unveiled the X7 Ultra large-model chip designed for AI inference, featuring a super-large-core NPU and, for the first time in China, a mass-produced Armv9 automotive-grade CPU, a semiconductor trade-intelligence report said.
The chip's NPU achieves 98.7% hardware parallelism utilization, cuts memory-bandwidth consumption by about 80% versus a flagship competitor and improves energy efficiency by roughly 100%, with dedicated acceleration for transformer operators such as Layernorm, Softmax and Transpose.
XHEART said the architecture supports a tenfold increase in deployable model parameter scale, while the Armv9 automotive CPU delivers about 20% higher performance and 10% better efficiency, integrating a new SVE2 vector-processing engine.
The launch highlights how China's AI-chip designers are pairing domestic large-model inference silicon with advanced automotive-grade compute to serve both edge-AI and intelligent-driving workloads.
Source: semiconductor trade-intelligence report / XHEART launch (paraphrased and translated from Chinese industry-media reporting; reprinted for informational purposes with attribution).
