Training Mamba model with the same number of parameters is way slower than its CNN counter part.

[pooya-mohammadi]

I train a CNN model with say 40,000,000 parameters it's way faster than a mamba model with 721,079 params.
Is it usual or something wrong on my side?
I tried it on a desktop 4090 and a server with A100. The output is the same for both of them.