On Mon, Oct 29, 2012 at 3:22 PM, Antoine Pitrou
On Mon, 29 Oct 2012 09:56:57 -0400 Brett Cannon
wrote: To see if the bad iterative_count and threaded_count results were consistently bad, I ran the benchmark suite on my MacBook Pro to see how "reliable" the benchmarks were. The output is below.
Basically 6 benchmarks (regex_effbot, queens, startup_nosite, iterative_count, threaded_count, and telco) had a variance of more than 15% performance between my 2 computers, although queens, iterative_count, and threaded_count were the only ones that swung between neutral/good to bad depending on the machine (the rest either want from bad to very bad, or very good to more very good).
This is using different compilers on the 2 computers, right?
Yes: gcc 4.6.3 on Linux and Clang 3.1 on OS X.