Uploaded October 2025 | Updated September 2026, 2 weeks ago
Once again revisiting ball dropping with my now more efficient approach for dealing with hundreds of millions of balls. Here 2^27 balls (approx 130 millions) are dropped on the minima of the interior of an ellipse, revealing some quite bizarre chaotic, yet symmetric, collective behaviour of the balls.
PyTorch was used for all GPU related calculations on my M3 MAX 64GB machine (with its GPU like MPS). 560 000 steps were calculated, and only each 50th frame was saved to disk (approx 11 000 frames in total) to smoothly overcome fringing/artefacts by the temporal discretisation of the ball paths. Yet the calculation only took 8 hours (step time = 0.05 s on average, graphics rendering included).
Beautiful music by @gpcbass . The song is called Flowers. An alternative version, by Retrobites, can be seen here: youtube.com/watch?v=VjHtFWKmRE4
Once again revisiting ball dropping with my now more efficient approach for dealing with hundreds of millions of balls. Here 2^27 balls (approx 130 millions) are dropped on the minima of the interior of an ellipse, revealing some quite bizarre chaotic, yet symmetric, collective behaviour of the balls.
PyTorch was used for all GPU related calculations on my M3 MAX 64GB machine (with its GPU like MPS). 560 000 steps were calculated, and only each 50th frame was saved to disk (approx 11 000 frames in total) to smoothly overcome fringing/artefacts by the temporal discretisation of the ball paths. Yet the calculation only took 8 hours (step time = 0.05 s on average, graphics rendering included).
Beautiful music by @gpcbass . The song is called Flowers. An alternative version, by Retrobites, can be seen here: youtube.com/watch?v=VjHtFWKmRE4
