Recently, the public benchmark test for training machine learning neural networks, MLPerf, has introduced new tests specifically for training large language models and generating images based on text. Companies like NVIDIA, Intel, and Google have invested in massive systems to participate in these tests, highlighting the competitive nature of the generative AI training field. NVIDIA's 10,000 GPU supercomputer completed GPT-3 training in the shortest time, but Intel saw significant performance improvements after enabling 8-bit floating-point calculations. NVIDIA and Microsoft's systems are in fierce competition for GPT-3 training. Training generative AI requires immense computational power, and in the future, companies will continue to expand the scale of their supercomputers to maintain their lead.