Method A) would be to install the fragment generating machinery yourself. I believe Rosetta3.3 includes a very useful demo of the actual fragment generating machinery which works very well (from experience) (rosetta_demos/fragment_picker/); unfortunately you still have to get results from a bunch of third-party secondary structure prediction programs to get the input files for the fragment picker, I can’t help you with that part.
Method would be to break your problem up into smaller bits if at all possible. abinitio scales very poorly to 1000 residues. mr_rosetta may do better.
Method C); you should be able to do the fragment generation shotgun style. Break your 1000 residues into two groups, one of the first 600 residues and one of the last 600 residues (with 200 residues in both sets). Robetta can handle that. Then, merge the files (throwing out the last 100 residues from the first group, and the first 100 from the second group, because they are falsely near the edge) and do some scripting to get the fragment file numbering to work out right. Does that make sense?