I'm reading about memory efficiency optimizations for FWI and RTM, and I found many papers that adopt the checkpoint or checkpoint + compression instead of using boundary checkpointing.

Salamanca et al (2018) show that 3D RTM can benefit from boundary saving and claimed that it is way more efficient in terms of execution time and storage necessities. Is there any counterargument to it? If not, why are checkpointing and compression still being used?

Salamnca et al (2018). Comparative analysis of 3D RTM Implementation Strategies for an Efficient Use of Memory in a Single GPU.