Surpassing 10T scale for fine-grained data access to storage

Vector database indexing and search, and eCommerce usage of GNNs would be constrained in size if they had to fit in the few terabytes of storage of a single node. Data set sizes are trending beyond what will fit in the combined memory capacity of a whole rack of GPUs and CPUs. To meet the needs of both single-node and rack-scale enterprise and hyperscaler customers, there's a need to scale up to storage, rather than just scaling out more. We'll show fresh results for WholeGraph and cuVS at 10-trillion scale, along with the ecosystem prototypes that light the path toward commercial deployment.