What is the cost of partitioning cells into black/red? Is it significantly smaller than the costs of synchronization?

What are the biggest differences between the Data-parallel abstraction and the shared address space? These two models seem like the same exact thing to me.

@evs The difference that in the data-parallel model, it's the software that handles the assignment of work between our computational resources. By contrast, the shared address space model requires us as the programmer to specify much more explicitly which threads are responsible for which tasks. I think of data-parallel as a bit cleaner and more relying on the interface that's given to us, and shared address space as being a bit closer to the hardware and having more control over the implementation.

When we complete the computation of the red cells, the black cells would still have their old values. But when we now work on the black cells, they use the new red values instead of the old ones (which was the case in the original problem statement). I am just wondering if this would cause a difference in the outcome?

It seems like the data-parallel model would be more widespread due to its ease of use.