> GPUs (7) are the common choice of hardware for training large language models. However, there is a global shortage of GPUs due to the large investment from ma

> GPUs (7) are the common choice of hardware for training large language models. However, there is a global shortage of GPUs due to the large investment from many countries to train LLMs. Under such circumstances, it is important to show that large language models can be trained using Fugaku, which uses CPUs instead of GPUs. The CPUs used in Fugaku are Japanese CPUs manufactured by Fujitsu, and play an important role in terms of revitalizing Japanese semiconductor technology.