> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# How long does graph materialization take?

> Use measured index-build costs and your data and hardware to estimate preparation time

Materialization time depends on graph size and structure, source read performance, and CPU, RAM, and local SSD capacity.

The published synthetic commerce benchmark built an index for **50 billion nodes and 175.5 billion edges in 49.8 minutes**. It used one AWS i4i.32xlarge with 128 vCPUs, 1 TiB RAM, and 30 TB NVMe storage. Peak RAM during the build was **636 GiB**.

That measurement covers index preparation. Training-table generation, model training, and prediction are separate. The benchmark's neighborhood-generation timings exclude the index build.

See [preparation costs and test conditions](/kumo-engine/resources/benchmarks#context-generation-and-preparation). Measure a representative build on your own hardware to estimate production runtime.