GCC 9 Compiler Tuning Benchmarks On Intel Skylake AVX-512

The C-Ray multi-threaded ray-tracer always craves a good compiler tuning.

These results basically jive with our past experiences in Intel Skylake-AVX512 compiler tuning and for tweaking the CFLAGS/CXXFLAGS in general for achieving greater performance. For most workloads, it pays off if you are satisfied with building out your own software packages.

This is an overall look at the performance of the 71 benchmarks when taking their geometric mean. Hitting at least "-O2" generally captures most of the performance potential, opting for "skylake-avx512" over "skylake" on capable AVX-512 CPUs did tend to really help out select workloads but not across the board, the benefits of link-time optimization (LTO) really come down to the application being tested, and opting for the most aggressive "-Ofast" mode helped out select cases. Those wanting to dig into this data more can find all of the numbers on OpenBenchmarking.org.

The GCC 9 stable release (GCC 9.1) should be out in the next few weeks.