I How much number crunching is necessary in some simulations?

[lpetrich]

I've seen simulations like these, simulations that were done in *very* gory detail, and simulations that apparently required a large number of CPU cycles and a large amount of memory space.

• Cosmological structure formation and galaxy formation
• Collisions of galaxies
• Formation of planetary systems and collisions of planets
• Inspiral of black holes or neutron stars

Does anyone have any more details on any of that? Like what computers that they used and how long that they needed for a run.

 

[phyzguy]

Almost all of these simulations are done on massively parallel systems of computers. Each computer is similar to the one in your laptop, but there are many thousands of them connected together. The problem is then split into pieces so that each computer is just solving a piece of the problem. In the cosmological structure simulations, each computer is working on a small volume of space, and the computers send information back and forth about what happens on the boundaries so the simulation can be stitched together into the whole volume. The Illustris Simulation, for example, states that "The largest was run on 8,192 compute cores, and took 19 million CPU hours". If you divide this out, it must have taken about 3 months of elapsed time running on 8192 computers.