On November 19, Zhongcheng Hualong Computer Technology Co., Ltd. held a press conference for intelligent computing products, officially launching the HL series of fully domestic high-end AI training and inference chips along with a full-stack intelligent computing product line. Zhongcheng Hualong focuses on core technology, reconstructing the efficiency and cost boundaries of AI computing power, accelerating the large-scale implementation of AI with inclusive computing power. The launched HL series AI chips boast advantages such as strong performance, large capacity, high energy efficiency ratio, and cost-effectiveness. The HL100, which made its debut, is a fully domestic AI training and inference chip equipped with a self-developed next-generation GPGPU+NPU integrated architecture, covering all computing scenario acceleration units, supporting full precision computing (FP32/TF32/FP16/BF16/INT8), and deeply supporting mixed precision computing. The FP16 computing power reaches 256 TFLOPS, and it is equipped with a cost-effective LPDDR5 video memory, supporting an ultra-large capacity of 128GB per chip, which is 1.33 times the memory capacity of a certain foreign AI chip. Additionally, the HL100 supports deployment from single-machine multi-card to multi-machine multi-node clusters, allowing flexible expansion to a scale of thousands of cards. The energy efficiency ratio of the Zhongcheng Hualong HL100 reaches 3.41 TFLOPS/W, providing 8 times the computing power of a certain foreign AI chip under the same power consumption; under the same computing power, its total cost of ownership (TCO) is 1/4 of that of a certain foreign AI chip. Currently, the HL100 can comprehensively cover mainstream large models and traditional models, accelerating implementation based on deep optimization of operators, natively supporting the PyTorch framework to significantly shorten the model adaptation cycle, and achieving flexible computing power allocation through the vNPU virtualization technology architecture. According to the information disclosed at the press conference, the next-generation products HL200, HL200Pro, and HL400 AI chips will be launched successively. The HL series fully domestic AI chips natively support FP8/FP4, with performance on par with international mainstream AI chips, fully meeting the training and inference needs of the next generation of generative AI. As a key support point in Zhongcheng Hualong’s “chip + complete machine + solution” full-stack product layout, the HL series AI chips represent a new choice for high cost-effective computing power that is independently controllable in the domestic AI industry, injecting new momentum into the inclusive domestic computing power.