Quick start
-
Enter count and dims
e.g. 100k vectors at 1536 dims (text-embedding-3-small).
-
Pick dtype
float32, float16, or int8 sets bytes per value.
Embedding Storage Calculator
Quickly estimate dense embedding payload size when planning RAG or vector DB capacity (index excluded).
Privacy: processed locally, never uploaded.
↓ Paste in the input area below to see results instantly
Estimate raw vector storage from count, dimensions, and dtype (index overhead excluded).
Bytes
614,400,000
MB
585.94
GB
0.5722
Notes
Storage note
Covers dense vector payload only. Vector DB indexes, metadata, and replicas add more disk use.
Quickly estimate dense embedding payload size when planning RAG or vector DB capacity (index excluded).
Index overhead
HNSW and IVF indexes often exceed raw vectors; this tool counts payload only.
Typical Workflow
When deploying a RAG system, developers typically first identify the embedding model (e.g., text-embedding-3-large with 3072 dimensions) and estimate the number of document chunks. By inputting these three key parameters (count, dimensions, dtype), this tool instantly calculates raw storage requirements, aiding in server configuration or cloud storage planning.
Example: 1 million float32 vectors with 768 dimensions occupy ~3GB. Switching to float16 halves the size but may lose precision. The tool allows real-time parameter adjustments for comparing results, ideal for validating storage costs during model selection.
Examples
Example
Input
100000 vectors, 1536 dims, float32
Output
~586 MB
FAQ
Includes metadata?
No document IDs, payloads, or other metadata fields.
Why does my actual database storage exceed the tool's estimate?
Vector databases require extra space for index structures (e.g., HNSW graphs), document metadata, and compression overhead. Milvus/Pinecone typically need 1.3-2x raw data size. This tool focuses on raw calculation – adjust budgets per your database's documentation.