Hello, I am setting up a new GPU cluster and noticed that the NUMA affinities when running nvidia-smi topo -m are uneven
GPU0->0
GPU1->0
GPU2->0
GPU3->1
GPU4->2
GPU5->2
GPU6->2
GPU7->3
I understand that this is due to the internal hardware of my motherboard, but I am concerned that the uneven distribution of numa nodes may affect performance. Does anyone have info on this? Thank you!