I'm wondering if the interconnect between GPUs in this HPC which seems like a more traditional supercomputer good enough for LLM training?