Rosetta is probably the most widely used program for this problem. It performs very well in many macromolecular modeling problems including Ab initio. You get the software from http://www.rosettacommons.org/. RosettaCommons also has tutorials, examples, and a user forum. The methods have been described in dozens of journal articles.
Not sure if the currently available abinitio methods are mature enough to model a full-length protein of 100 residues. I would suggest you to decompose your sequence into secondary structure elements using tools like PSIPRED. Then identify all loop residues and model loops separately using loop modelling approaches and model the remaining residues using ab initio approach.
You may also interested in a web server based on the Rosetta method. Robetta: Full-chain protein structure prediction server. It uses a hybrid approach to model query sequence. Detailed help page is available here.