Zhongcheng Hualong HL200 Inference Chip Released: Energy Efficiency 5.12 TFLOPS/W, Cluster Scalable up to 10,000 Cards

Zhongcheng Hualong has officially launched the HL200 inference chip and the super-node intelligent computing cluster solution. The chip natively supports FP4 and FP8 low-precision inference, with a single-card FP4 computing power of 4P and an energy efficiency of 5.12 TFLOPS/W.

Demand for energy efficiency and low-precision computing in the current AI inference market is growing. The HL200 natively supports FP4 and FP8 formats, helping to reduce computing costs and power consumption while maintaining performance, aligning with the industry trend towards high-efficiency inference.

The accompanying intelligent computing cluster solution supports large-scale expansion, with a single cabinet capable of interconnecting 64 GPUs, and horizontally scalable up to a scale of 10,000 cards. This high scalability design meets the large-scale inference deployment needs following large model training, providing a hardware foundation for building high-performance intelligent computing centers.

With the popularity of generative AI applications, the demand for inference computing power has surged. Domestic chip manufacturers are actively positioning themselves in the inference market. The products released by Zhongcheng Hualong this time further enrich the choices of domestic AI computing chips, helping to promote the maturity of the relevant industry chain.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.