AMD Bulldozer With GCC, Open64, LLVM/Clang Compilers
With the C-Ray benchmark, which is a multi-threaded ray-tracing test and one of my favorite tests overall for pounding multi-core systems, the clear winner was the Open64 compilers. Open64 was significantly faster than LLVM/Clang on the AMD Bulldozer platform. It's not a huge surprise as while Open64 isn't nearly as common as GCC, it's AMD's preferred compiler for their systems and does offer the most right now in terms of Bulldozer optimizations.
LLVM/Clang on the eight-thread Buldozer is faster than GCC 4.6 by a nice margin, but GCC 4.7 manages to speed-up by a great deal to push it ahead.
LLVM/Clang is significantly slower than GCC with the Smallpt path-tracing benchmark. Open64 could not successfully build this test.
For the single-threaded POV-Ray ray-tracing benchmark, LLVM/Clang 3.0 SVN was faster than GCC and Open64.