[dcohenp]

I'm a bit surprised at the "computational benchmarks" [1] though. Can anyone confirm these are indeed CPU-bound, and not somehow GPU-assisted? If so, how can the underlying kernel possibly make such a difference? Scheduling? (if so, what kind of load was the machine running?) Clearly there must be something more to it.

[1] http://www.phoronix.com/scan.php?page=article&item=osx10_ubu...

[teamhappy]

PolyBench/C is open source. Straight forward matrix multiplications in C. The code looks exactly like you'd expect it to look.

See the description for implementation details: http://web.cse.ohio-state.edu/~pouchet/software/polybench/#d...

[chromakode]

Perhaps power management is somewhat to blame here? I would be curious to see how power consumption compared while running the benchmarks. It's well known that OSX gets much better battery life than Linux on Apple hardware. I wonder if the CPU was being throttled to conserve power or thermal output.

[teamhappy]

> It's well known that OS X gets much better battery life than Linux on Apple hardware.

I couldn't find any information to back that up. Do you have a link to a benchmark?

I don't think kernels throttle CPUs — CPUs do that themselves already. I could imagine Apple doing something like OpenBSD in that they favour security over performance for some operations, but that seems unlikely. IOKit might be to blame as well, given that it's written in embedded C++ instead of C.

All things considered, I think it's highly likely that, while OS X is quite shiny on the outside, the underlying OS isn't as modern as some of the others. Windows and OpenBSD have better security features and now Linux apparently has better performance.

[dcohenp]

Well, but I guess theoretically it could be doing stuff like setting lower thermal thresholds for throttling. So perhaps not exactly a function of scheduling, but getting the CPU to make itself slower.

Still, this to me is the most interesting result from that set, and I'd really love it for someone to dig into it.