You are storing 1,000 embeddings in a VECTOR column, each with 256 dimensions using FLOAT32. What is the approximate size of the data on disk?

To determine the size of the data on disk when storing embeddings in a VECTOR column, it’s essential to understand the calculation behind the size of these embeddings.

Each embedding consists of 256 dimensions, and since each dimension is represented in FLOAT32, each dimension will take up 4 bytes of storage (as FLOAT32 is a 32-bit floating point). Therefore, for one embedding, the memory requirement would be calculated as follows:

256 dimensions * 4 bytes per dimension = 1024 bytes, or 1 KB per embedding.

Since you are storing 1,000 embeddings, the total size in bytes can be calculated as:

1 KB per embedding * 1,000 embeddings = 1,000 KB.

To convert this to megabytes, you divide by 1024 (as there are 1024 KB in a MB):

1,000 KB / 1024 KB per MB ≈ 0.976 MB, which rounds to approximately 1 MB.

Thus, storing 1,000 embeddings with 256 dimensions using FLOAT32 takes up about 1 MB of space on disk, confirming the choice provided.