Currently, managed memory is the preferred approach for sampling in resource-constrained environments, but it has a number of drawbacks, including large page sizes. UVA is already used by DGL in its GPU-enabled sampling approach; it would be good to see how it performs within cuGraph.
Currently, managed memory is the preferred approach for sampling in resource-constrained environments, but it has a number of drawbacks, including large page sizes. UVA is already used by DGL in its GPU-enabled sampling approach; it would be good to see how it performs within cuGraph.
See: https://github.com/rapidsai/rmm/pull/1392