Vector Search easy
You are estimating memory for 10 million embeddings, each with 1,536 dimensions stored as fp32. Roughly how much RAM do the raw vectors need?
- A About 6 GB, since each vector takes roughly 600 bytes in fp32.
- B About 60 GB, since each vector takes roughly 6 KB in fp32.
- C About 600 GB, since each vector takes roughly 60 KB in fp32.
- D About 15 GB, since each dimension takes roughly one byte in fp32.
Answer: B. About 60 GB, since each vector takes roughly 6 KB in fp32.
Each fp32 value is 4 bytes, so 1,536 x 4 is about 6 KB per vector, and 10 million vectors come to about 60 GB before any HNSW overhead (often 1.5x to 2x more). Option D describes int8 quantization, which does cut memory by 4x but costs some recall. That option is only right if you have chosen to quantize.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- You are building an IVF index on 20 million vectors. A teammate asks why the build needs a 'training' step before any vectors can be added. What is the reason?
- An HNSW index serves a marketplace where about 30% of listings are updated or removed each week. Each update is applied as a delete plus an insert, and deletes mark tombstones. Over a month, recall@10 drifts from 97% to 89% and p99 latency rises at the same ef_search. What is happening, and what should you do?
- Users delete documents in your primary database, but the deleted documents keep appearing in vector search results for hours. The vector index is rebuilt nightly. What is the best fix?
- Your embedding model outputs unnormalized vectors whose magnitudes vary widely. A teammate wants to switch the index from cosine similarity to dot product because it is cheaper to compute. What is the consequence?
- You are setting an SLO for a new ANN-backed recommendation search over 50 million vectors. A product manager asks for 100% recall@10 with single-digit millisecond latency. What is a realistic response?
- A B2B SaaS app serves 500 customers, and every query searches only one customer's documents. One large customer runs heavy batch queries that slow down search for everyone. Which sharding approach fits best?