That may might not be true. Assuming nothing travels faster than the speed of light. Next assume your calculating all information in some unit of space. Now for small increases in time you can ignore everything further than that time * speed of light. Now, repeat that process for each point in space for 1 tick. Then start over.
Think of it like simulating a really large map of 'game of life', you can chunk them into separate threads and only pass information about the overlap after each calculation.
Since each field component in the voxel depends only on its immediate neighbours, you can perform massively parallel simulations on GPUs.
I've worked on this stuff, it's pretty neat. Discretizing continuous differential equations onto a voxel lattice is how a lot of things these days are done, including seismic and medical imaging.
Think of it like simulating a really large map of 'game of life', you can chunk them into separate threads and only pass information about the overlap after each calculation.