Six standalone experiments explore the path from particle initialisation to GPU computation and native rendering. C++ generates positions, velocities and masses; Objective-C++ manages Metal resources; compute kernels calculate softened all-pairs gravitational acceleration and update motion. Scenarios include clouds, spiral arms and counter-rotating clusters.
The suite preserves baseline and tiled force calculations. Tiled variants cooperatively stage particle data in threadgroup memory, changing memory access while retaining quadratic computational complexity. MetalKit viewers render directly from simulation position buffers, while console harnesses report completed-step timing summaries after warm-up.
This is an educational simulation and performance exploration. Reproducible baseline comparisons, CPU-reference correctness checks and stability measurements remain future work. The latest audit could not run the simulations because the local Metal Toolchain was unavailable, so no measured speedup or scientific validation is claimed.