GCC vs. Clang Compiler Performance On NVIDIA Xavier's Carmel ARMv8 Cores

While years ago Clang would almost always compile code faster than GCC, more recent releases of the GNU Compiler Collection have put up a tougher fight and with the PHP compilation test actually was faster than Clang 6.0 by a considerable margin.

Similar to on x86_64, C-Ray saw much faster performance when built by GCC.

Primesieve and FLAC were a toss-up for performance.

The Redis performance on the NVIDIA Tegra Xavier SoC was slightly faster with the Clang-generated binaries.

The Apache web-server performance was similar between the compilers.

When looking at the wins and losses across all of the tests, GCC on Ubuntu 18.04 LTS won 55% of the tests to Clang winning 45% of the tests with these stock compilers from the current Ubuntu Long Term Support release on this Tegra Xavier SoC. When looking at the geometric mean of all the test results, it was also a pretty even fight between these compilers on this octal-core ARMv8 SoC. The results aren't all that surprising considering the many ARM vendors relying upon Clang in production and overall a great deal of interest from LLVM within the mobile/embedded space has resulted in a lot of tuning.

With the GCC 9 and Clang 8 releases around the corner, larger ARMv8 compiler benchmark comparisons will be coming up soon on Phoronix from different 64-bit ARM developer boards to complement all of our ongoing and frequent Linux x86_64 compiler benchmarks.