Hi,
I got a single bridge connecting 2 A100 40gbs in a supermicro AS -4125GS-TNRT across the GPUs’ first nvlink slot, I have the cuda toolkit, cuda 12.3, and 545 driver installed.
But nvidia-smi topo and nvlink -s both show the absence of nvlink
Is this because I’m using only 1 bridge instead of 3? Or is there another issue.
Bridge is seated fully as per the image.
Thanks
Hi, I was looking for the answer to this question seperately and found this while I was looking. Since I don’t see any replies here I thought I’d just let you know that it is not supported. the only supported arrangement appears to be 3 bridges between 2 cards.
source a100 80GB product brief: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/PB-10577-001_v02.pdf
2 Likes
Thanks for the reply, I also found a dell rack information sheet that says you need all 3 bridges.
Quite unfortunate