While there probably isn’t any *major* changes to any of the protocols, there’s typically small bug fixes and small improvements to feature sets. For example, relax (and other protocols) has some extra options in 3.4 versus 3.3 (but probably nothing you’d use in a general case), and there’s extra options for tweaking/controlling scoring and controlling running (but again, probably nothing you’d employ in the general case).
You can usually get a good impression as to whether something significant changed by comparing the documentation for the application between the two versions. Major changes should get mentioned in the documentation, although minor updates probably won’t.
If you already have Rosetta 3.3 built and working, and you’re happy with the results, there likely isn’t any compelling reason to update just for the sake of updating. Although if you’re still really wondering about it, the best thing to do is run test cases side-by-side on the two version, and compare the performance.
There will likely be some updates (along with published papers) in the near future regarding ab initio and comparative modeling structure prediction (it is a CASP year, after all), but none of that is yet “released” with Rosetta 3.4.
.
