That's not the distinction I'm looking for. I'm looking for a better term to describe the throughput the application would achieve if all non-memory constraints were removed. In the context of the article, the author is comparing the ratio of the "effective/observed bandwidth" and the "theoretical peak bandwidth", noting that the ratio is large, and concluding that he is not constrained by memory bandwidth. I don't know that this is a reasonable conclusion.

I'm looking for a different denominator, which is also theoretical. Maybe call it the "theoretically achievable bandwidth", which takes into account all the details of the requests. If the "observed bandwidth" equals the "theoretically achievable bandwidth", your only path to improvement is to change your data layout or increase the parallelism of your requests. If the "observed bandwidth" is less than this theoretical, there should still be room for implementation optimizations.

Falling short of the "theoretical peak bandwidth" (even by a lot) doesn't tell you which of these is the case.