In order to test the self-gravity implementation in CHOLLA, I ran a adiabatic simulation of an ideal gas collapsing under it’s own gravity.
The box size is , the density in the box is and for the spherical region of there is an over-density of . Initial velocities are uniform and pressure is uniform .
The next figures show density and potential center slices. The black circle corresponds to the analytical value for the radius of a uniform pressure-less system collapsing only under gravity.
Timing for several grid configurations
All times are in seconds and correspond to the average of the first 20 time-steps ( not counting the first two time-steps )
For the XSTREAM system there was no debug queue, this makes any development to be very tedious, and I wasn’t able to run more than 128 GPUs.
NOTE: This was the implementation of the poisson solver that wasn’t optimized for solving real-complex-real FFTs, also there was another optimization that avoided the last transposition for the forward FFT saving communication time.