Relaxed Memory Model Zoo

A living survey of published weak memory models

The Relaxed Memory Model Zoo is a living survey of weak memory models: 107 models spanning 47 years — hardware, programming language, distributed and transactional — and 161 relations between them, ordered by the inclusion of the behaviours they allow. A time slider makes the chronological development of the field explorable, and the dataset behind the map is machine-readable.

A recording of the Zoo in use: the map of memory-model nodes ordered by strength, with every arrow descending from a stronger model to a weaker one; selecting C11 opens its definition, tags and neighbours; an edge's evidence expands to the herd7 litmus test that separates the two models; the category, cat-specifiability and property filters narrow the map; the year slider sweeps from 1979 to 2026; and search matches names, descriptions and references.
The Zoo's map, from overview to the litmus test behind a single edge. [fig1]

Every model and every edge carries provenance naming the original publication, and every ordering claim states how it is known: a litmus test in the repository that runs and produces the verdict, a mechanical comparison of the two models with kater or memalloy, an argument by construction, or a citation to the primary source. 85 herd7 checks against a pinned herdtools7 and 6 kater containment proofs run on every push, and an edge that contradicts the transitive closure of the others fails the consistency gate before it can ship.

The dataset lives in rmm-zoo-dataset, which is where corrections and additions go. If the Zoo is useful to you, follow the repository on GitHub — that is the signal I go by. Should it prove useful to the community, the Zoo will move off this subdomain and onto a domain of its own.