A new register allocator (LRA) has been added, improving code performance.

A fast memory error detector called AddressSanitizer has been added.

A fast data race detector called ThreadSanitizer has been added.

Extremely large functions can now be compiled faster using less memory.

A new general optimization level has been introduced.

GCC diagnostic messages now highlight the exact problem in the source code.

Various new optimization options have been added.

DWARF4 is now used as the default debug format.

GCC now fully implements the C++11 language standard.

C++11 library support has been extended though is still experimental.

GCC now supports dynamic initialization of thread-local variables.

Support has been added for Intel FXSR, XSAVE, and XSAVEOPT instructions.

New built-in functions added to detect run-time Intel CPU Type and ISA.

Intel function multi-versioning support added.

Intel RTM/HLE intrinsics, built-ins, and code generation have been added.

Transactions (the -fgnu-tm option) can now be run using Intel TSX extensions.

Support for AMD family 15h and 16h processors has been added.

Various Fortran changes have been included.

Improved and expanded support for Python scripting.

Improved handling of C++ debugee executables.

Improved inferior control commands.

Improved support for ambiguous line specifications.

Improved tracepoint support.

Multi-program debugging.

The new gold linker, which is smaller and faster than ld. Note that gold is not the default linker and must be explicitly enabled by using the alternatives command.

Support for link-time optimization (LTO) in conjunction with GCC.

Support for build-IDs, unique numbers to identify executables.

Support for the IFUNC and UNIQUE symbols that are used by glibc to improve performance. Due to dependencies on a particular version of the glibc library, these symbols are only available on Red Hat Enterprise Linux 6.

Compressed debug sections for smaller debug info files.