[Zardoz84]

I got a speedup of around 13,32 on a 16 core Opteron machine a few years ago, using a simple parallel foreach on a N Body problem simulator. I didn't do anything special for it.

I thought using SIMD to accelerate it more on the integrator and acceleration calculation, inside of the parallel for.

[chrisseaton]

Unless you are creating or destroying bodies then your n-body implementation won't need to allocate any memory will it? So is it relevant to what he's talking about?