It is difficult to predict the exact run time for RAxML, as it depends on many factors that you didn't disclose here. For example, 723 protein sequences is a not a huge alignment, but we don't know the alignment width. If you have 5000 aligned columns in your dataset, that will take longer than an alignment that is 150 columns wide. Next, how many threads are you using? The speed-up is not exactly linear, but more threads means less running time. I have no experience with how Windows handles RAxML, but anything running under a GUI could be less than optimal, especially if you are using computer for other purposes in parallel.
To give you a point of reference: under Linux, an alignment of ~2000 sequences that is ~3500 residues wide takes less than a day on 30 CPUs. To get an idea how far along the run is, count the number of lines in a file that begins with
RAxML_bootstrap. as the run will be done when you have 100 lines (100 trees) in that file. On Linux, that would be as simple as:
wc -l RAxML_bootstrap.*