[Sharlin]

The standard Floyd–Warshall is fairly easily parallelizable. I wonder how fast you could solve this problem with today's GPUs, and whether a(6) might be attainable in some reasonable time.