There are several factors, and several trade-offs, to consider here.
The server machine is usually stronger than the client machine. OTOH, there are usually far more clients than servers. Thus, when #clients exceeds a certain threshold, client-side processing is faster (the server will have to process computations from all clients which will outweigh its stronger processor).
But, if the processing is bounded mainly by network bandwidth, that is: most processing time is spent on downloading stuff, and the downloaded material is relatively stable, then it will be faster to download once into the server and do the processing there.
Finally, if the results of the computations do not change from one client to another, then - again - it makes more sense to compute it, once-and-for-all, at the server.