SpiceDB load tests will prove that SpiceDB can meet performance targets and determine how much hardware is needed to achieve them. The performance characteristics of SpiceDB are nuanced, and relationship data distribution significantly impacts computational cost. Seeding relationships for a realistic load test is essential, taking into account the cardinality of relationships and their impact on fanout. Checks, lookups, and writes can have significant impacts on performance, with checks being less computationally expensive than lookups. The quantization interval setting determines how long cached results will live, and consistency has a significant effect on cache utilization. Load generation tooling, such as Thumper, can help distribute requests in a realistic manner. Monitoring SpiceDB metrics is crucial to fine-tune the load test, and scaling hardware up or out may be necessary depending on CPU utilization. The Authzed team offers assistance with schema review, script creation, trial access to SpiceDB Dedicated, and consultation services.