For large bodies/deep water, it still isn’t practical to simulate the entire volume as particles, even with any acceleration techniques. If the surface can be treated as a heightfield it can be much more efficient.
I found this work really nice:
http://visualcomputing.ist.ac.at/publications/2017/WWP/
and here a slightly older simpler approach:
For splashing/breaking waves, heightfields alone aren’t enough, but there has also been some interesting work on combining grid based volumetric simulation with particles for the surface layer.