Google Gemini 1.0 outperforms GPT-4 in terms of performance, with its TPU v4 and TPU v5e utilizing optical circuit switching technology, significantly enhancing the speed of information transmission. The concept of MFU (Model FLOPs Utilization) is used to assess the load capacity and operational efficiency of TPU v4, achieving an MFU of 44-56% in benchmark tests. Gemini supports multimodal models, capable of handling complex, diverse, and unstructured data, with superior hardware performance compared to other models.