I've already delivered GCC vs. LLVM Clang compiler benchmarks on Intel's Core i7 4770K "Haswell" platform and tested the "core-avx2" optimizations offered by the latest compilers. That previous testing was done from the stable releases of LLVM Clang 3.1/3.2 and GCC 4.7/4.8 releases, but looking ahead, here's some benchmarks of the latest GCC 4.9 development snapshot.
GCC 4.8.1 stable was benchmarked against the GCC 4.9 2013-06-23 snapshot to see if there's any other performance changes yet for this next compiler release. Version 4.9 of the GNU Compiler Collection isn't expected until 2014 with the 4.8 series having just been stabilized in March. Among the changes already found in GCC 4.9 is colorizing of GCC diagnostics, many C++1y features have been implemented, and other changes have taken place.
Anyhow, the quick benchmarks between GCC 4.8.1 and GCC 4.9.0 2013-06-23 from the Core i7 4770K system can be found on OpenBenchmarking.org within 1306266-SO-GCC49INTE20. The CFLAGS/CXXFLAGS during testing were set to -O3 and -march=core-avx2 for hitting all of the Intel Haswell CPU instruction set extensions.For many CPU benchmarks, GCC 4.9 doesn't (yet) deliver any major performance changes over the binaries yielded by GCC 4.8.1.
For some benchmarks though, the GCC 4.9 SVN code is already getting faster, at least for the high-end Intel Core i7 Haswell processor.
One of the test cases where GCC 4.9.0 yielded the biggest leap in performance was the Himeno Pressure Solver.