# Atlas Required Memory metric sizing > Source-anchored research for Atlas Required Memory metric sizing. Run status: COMPLETED. Parent: [Atlas Cluster Topology and Tier Economics](https://llms-explorer.com/tree/atlas-cluster-topology-and-tier-economics/) · 4 facets · 9 facts · page: https://llms-explorer.com/tree/atlas-required-memory-metric-sizing/ ## How it works - Required Memory describes vector index memory, not total node capacity. MongoDB says Vector Search needs memory for both the index and the JVM; on co-located deployments, mongod also consumes node memory, leaving only the remainder for Vector Search and mongot. Add those process and workload needs when translating an index estimate into a node size. (high confidence) — [source](https://www.mongodb.com/docs/vector-search/deployment/deployment-options.md) ## Measurements and reference values - MongoDB reports that automatic scalar and binary quantization reduce vector RAM use by about 3.75x and 24x, respectively, while full-fidelity vectors remain on disk for exact search or rescoring; its deployment guidance suggests storage-to-RAM ratios near 4:1 for scalar and 24:1 for binary quantization, so Required Memory alone does not size disk. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/vector-search/deployment/deployment-options.md) - OneUptime's 2026-03-31 monitoring guide lists SEARCH_REQUIRED_MEMORY_SIZE and suggests alerting above 80% of available memory; the MongoDB sizing pages reviewed here give no such threshold, so treat this as a third-party heuristic pending workload validation. (low confidence) (verified as of 2026-04-13) — [source](https://oneuptime.com/blog/post/2026-03-31-mongodb-how-to-monitor-atlas-search-index-size-and-performance/view) ## Comparisons and alternatives - Inference: per-vector payload math is a starting estimate rather than total RAM. MongoDB says vector space scales linearly with vector count and dimensionality and gives representation-specific examples; its index also contains metadata and an HNSW graph, while the original HNSW paper describes a multi-layer graph over nested subsets of stored elements. Do not treat raw embedding bytes as the complete Required Memory value. (medium confidence) — [source](https://www.mongodb.com/docs/vector-search/deployment/deployment-options.md) - Datadog's Atlas integration lists separate gauges for required memory bytes for Vector Search indexes, total index size on disk, and Search process resident memory. This supports using Required Memory as an index requirement and resident memory as observed process telemetry, rather than treating the two as interchangeable measurements. (high confidence) (verified as of 2026-10-14) — [source](https://docs.datadoghq.com/integrations/mongodb-atlas/) ## Facts and statements - For automatically quantized Vector Search indexes, Atlas's Size represents total stored data, including the HNSW graph and quantized vectors in memory plus full-fidelity vectors on disk; Required Memory estimates query-time index memory, so use Required Memory for RAM sizing and account for disk size separately. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/vector-search/deployment/deployment-options.md) - MongoDB staff described Required Memory as an approximate value for each vector index; Datadog documents the corresponding Atlas gauge as mongodb.atlas.search.index.stats.required.memory, measured in bytes for vector search indexes. (medium confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/community/forums/t/automatic-scalar-and-binary-quantization-now-supported/306377) - Validate the Required Memory estimate against runtime pressure: MongoDB recommends keeping Search Page Faults near zero, says sustained faults mean queries are going to disk and may justify scaling RAM, and recommends monitoring available and used System Memory with Search Index Size. (high confidence) (verified as of 2026-10-14) — [source](https://www.mongodb.com/docs/vector-search/benchmark/performance-recommendations.md) - The Atlas Search changelog lists a required-memory usage metric in the 22 September 2025 release; MongoDB staff had described an approximate per-index UI value on 29 April 2025, and the sources do not say whether those refer to the same metric exposure. (medium confidence) — [source](https://www.mongodb.com/docs/atlas/search-changelog.md)