Will triton use cuda core to do tl.dot when cuda core is faster？

[gujiewen]

I found triton always use tensor core to do matmul，so I was wondering will it use cuda core in some situations. Or it is only designed for tensor core.