Kicking things off with some molecular biology benchmarks, for HMMer the latest LLVM/Clang 3.2 release is still slower than GCC. GCC 4.8 wasn't any faster than GCC 4.7, but LLVM/Clang 3.2 is slightly slower still and wasn't improved between the 3.1 and 3.2 releases. (For the compiler flags noted, the -lm part can be ignored as for some cases LLVM/Clang 3.1 had problems with the -march=native mode on the Intel Ivy Bridge processor so the ordering was switched around.)
LLVM/Clang 3.2 is close behind the modern GCC releases tested on this Ivy Bridge processor.
For the new BLAKE2 crypto algorithm, the benchmark wouldn't build with LLVM/Clang 3.1 but it now works with LLVM/Clang 3.2. The Clang performance was shortly behind that of the GNU Compiler Collection.
LLVM/Clang secures a win when it comes down to the Monte Carlo test within SciMark. GCC 4.8 makes some performance gains over GCC 4.7 for this computational test, but LLVM/Clang 3.2 remains noticeably faster on the Intel processor.
There isn't much difference between LLVM/Clang and GCC for the Fast Fourier Transform workload.