I compared the performance of two operating system/compiler combinations running on identical hardware using the linpack.c benchmark.

On hpux-ia64 using gcc -O3 -o linpack linpack.c (matrix size 200) I got: 205598.441 KFLOPS Using the HP cc compiler I got worse results.

On ia64 (the OpenVMS-Itanium system) using cc/optimize linpack.c link linpack.obj run linpack.exe (matrix size 200) I got: 288766.598

The OpenVMS compiler on exactly the same harddware as HP-UX generated a 40.5% increase in throughput.