Layer 4 · self-hosting reality check
What it actually takes to self-host Qdrant
The docs say 1 GB. In practice you want Vectors × dimensions × 4 bytes, in RAM, plus overhead — 1M × 768d is roughly 3 GB. Here is the honest version — real requirements, real monthly cost, what you will be maintaining, and the one thing that catches people out.
Usually reached from Pinecone alternatives, where Qdrant is one of the picks.
| RAM — documented minimum | 1 GB |
|---|---|
| RAM — what it really needs | Vectors × dimensions × 4 bytes, in RAM, plus overhead — 1M × 768d is roughly 3 GB |
| CPU | 4 vCPU |
| Disk | Similar to the memory figure; on-disk mode trades speed for capacity |
| Monthly cost | $25–60/mo for a few million vectors, against Pinecone's $50/mo floor plus usage |
| Setup time | 30 minutes |
| How you install it | single Rust binary or container — genuinely one process, which is the appeal |
| Ongoing maintenance | Low. Snapshots for backup, and watch memory as the collection grows. |
| Where it stops scaling | Hundreds of millions of vectors with quantisation and sharding. Single-node handles far more than most RAG applications need. |
The thing that catches people out
Do the memory arithmetic before you index, because the failure mode is abrupt. Vectors live in RAM by default and 10M × 1536-dimension embeddings is about 60 GB — a number that surprises people who sized against document count. Enable on-disk storage or scalar quantisation early; retrofitting after an OOM is worse than planning for it.
When not to self-host Qdrant
You already run PostgreSQL and your corpus is modest. pgvector adds vector search to the database you have, with no new service to operate.
Every guide here carries this section. A site that only ever tells you to self-host is selling something — the useful answer is sometimes no.
Other Layer 4 self-hosting guides
- Self-hosting Ollama8 GB VRAM for a 7B model at usable speed; 24 GB for 30B-class
- Self-hosting vLLM24 GB VRAM minimum for useful production serving
- Self-hosting pgvector8 GB — the HNSW index wants to be resident
- Self-hosting LlamaIndex4 GB for the app; your vector store is the real cost
- Self-hosting faster-whisper5 GB VRAM for large-v3 in float16; 2 GB with int8
- Self-hosting Langfuse4 GB
Common questions
- How much RAM does Qdrant actually need?
- Vectors × dimensions × 4 bytes, in RAM, plus overhead — 1M × 768d is roughly 3 GB in practice. The documented minimum is 1 GB, which is the figure at which the process starts rather than the figure at which it works under real use. 4 vCPU alongside it.
- What does self-hosting Qdrant cost per month?
- $25–60/mo for a few million vectors, against Pinecone's $50/mo floor plus usage This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for low. Snapshots for backup, and watch memory as the collection grows.
- How long does it take to set up Qdrant?
- 30 minutes, via single Rust binary or container — genuinely one process, which is the appeal.
- When should I NOT self-host Qdrant?
- You already run PostgreSQL and your corpus is modest. pgvector adds vector search to the database you have, with no new service to operate.
- What is the most common mistake when self-hosting Qdrant?
- Do the memory arithmetic before you index, because the failure mode is abrupt. Vectors live in RAM by default and 10M × 1536-dimension embeddings is about 60 GB — a number that surprises people who sized against document count. Enable on-disk storage or scalar quantisation early; retrofitting after an OOM is worse than planning for it.