High-performance vector searches depend on vector indexes and impose requirements on instance memory. Use this page to find the minimum instance size for your vector dataset, based on its dimension count and row count.
The specifications below apply to in-memory indexes only. If you use an HGraph index withprecise_io_typeset toreader_io(a hybrid memory and disk index), the same instance can support more vector data.
Recommended minimum instance specifications
Find your dimension in the left column, then read across to the row-count range that matches your dataset. The table covers single-table scenarios. Test against your actual queries per second (QPS) and latency targets, and increase the instance size if needed.
| Vector dimension | Row count | Minimum instance |
|---|---|---|
| 128 | Less than 200 million | 32 CPU cores |
| 128 | 200 million to 400 million | 64 CPU cores |
| 128 | More than 400 million | 128 CPU cores or above |
| 256 | Less than 60 million | 32 CPU cores |
| 256 | 60 million to 120 million | 64 CPU cores |
| 256 | More than 120 million | 128 CPU cores or above |
| 512 | Less than 30 million | 32 CPU cores |
| 512 | 30 million to 64 million | 64 CPU cores |
| 512 | More than 64 million | 128 CPU cores or above |
| 768 | Less than 24 million | 32 CPU cores |
| 768 | 24 million to 48 million | 64 CPU cores |
| 768 | More than 48 million | 128 CPU cores or above |
| 1024 | Less than 16 million | 32 CPU cores |
| 1024 | 16 million to 32 million | 64 CPU cores |
| 1024 | More than 32 million | 128 CPU cores or above |
| 1536 | Less than 10 million | 32 CPU cores |
| 1536 | 10 million to 20 million | 64 CPU cores |
| 1536 | More than 20 million | 128 CPU cores or above |
Worked example: If you have 30 million 768-dimension vectors, the table shows a gap between the 24-million (32 cores) and 48-million (64 cores) thresholds. Start with 64 CPU cores and validate against your QPS and latency targets before scaling further.
For exact-match vector searches, no vector index is required. Scale your instance based on the table above using your row count and dimension.