Broadwell is your first 14 nm processor. Is the current Fortran compiler optimized for this new processor?
One would expect the AVX2 optimizations currently available and being improved in the current beta test compiler to carry over effectively.
If we take the Ivy Bridge introduction as a precedent, ifort didn't incorporate any changes which would make it imperative to re-compile applications built for Sandy Bridge. The new instructions for arbitrary precision integer and random number generation seem more likely to be used in libraries built with C compiler intrinsics than directly in Fortran. New graphics features might show up in OpenCL but not Fortran.
It seems likely to me that there would be more value in building applications which can run well on Haswell than in making them work only on Broadwell.
The question has a little of the flavor of the hype I hear around here from people who read the stock market tips and expect a new CPU model to make magic changes in the available markets. I regret that Haswell has been hitched to the success of new Windows versions. Not that I expect an ifort for Android to come out, but "it would be nice" if alternate OS like linux or even Mac weren't made difficult to use with new CPUs.
What Tim says is correct, but in addition to support for new instructions (Broadwell has a few), we also can have the compiler "tune" optimizations to favor a selected processor. Given what few new instructions Broadwell has, I wouldn't expect Fortran to make a lot of use of them. You can be sure that we will make sure Fortran compiles well for Broadwell as well as any future processors.