3 ms·
The lack of nvlink doesn't matter much for LLMs, which can work on distinct layers of the model on completely independent GPUs.
by cjbprime 3y ago
The lack of nvlink doesn't matter much for LLMs, which can work on distinct layers of the model on completely independent GPUs.