This weekend I set out to build a 64x64 LED matrix version of Conway’s Game of Life. For compute’s sake, the board would be wrap-around i.e. on a torus. I wanted each set of generations to run for as long as possible until stagnancy and reset, so I wrote an analysis script to determine what alive-percentage for a random fill would be optimal to keep the game going as long as possible.

The results surprised me – generations-until-stable (i.e. stagnant – oscillating, perhaps, but no meaningful movement) is fairly flat across a 20-60% range, flattening more and more as the board size increases. I ran a full 1%-100% sweep at 1000 samples per configuration, then a much longer one at 100k samples per configuration over 20%-50%. That data is here.

• 1k sample run raw data
• 100k sample run raw data

1k samples

100k samples