GCC 4.8 features improvements to link-time optimizations (LTO) with the LTO partitioning having been re-written for improved reliability and maintainability while fixing some link failures (see our GCC 4.7 LTO benchmarks). Aside from improving LTO, there's also improved inter-procedural optimizations. The interprocedural optimization improvements come from a new symbol table, improvements to the inline heuristicsm and better inlining decisions.
GCC 4.8 has also addressed some scalability bottlenecks within GCC's optimization passes that will benefit extremely large functions. There's also a new -ftree-partial-pre option for controlling partial redundancy elimination (PRE) optimizations.
Last but not least there is a new general optimization level for fast compilation but with superior debugging capabilities and still reasonable run-time performance.
Other new features to version 4.8 of the GNU Compiler Collection are talked about in this article.