3 ms·Towards Compute-Aware In-Switch Computing for LLMs on Multi-GPU Systems1 points by rbanffy 6mo ago