I assume it can be split between multiple GPUs, like LLMs can. Or hire an H100 for like $3/hr.