I believe the main advantage of the A100 is the memory bandwidth. Computationally the 4090 has a higher clock speed and more CUDA cores, so in that way it is faster.

So for this specific application it really depends on where the bottleneck is