Layer 4 · self-hosting reality check
What it actually takes to self-host LM Studio
The docs say 8 GB. In practice you want 16 GB system RAM, and VRAM is what actually decides your model. 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 OpenAI API (ChatGPT) alternatives, where LM Studio is one of the picks.
Wondering whether you need to at all? Is the OpenAI API free? — what the free tier actually allows, and where the wall is.
| RAM — documented minimum | 8 GB |
|---|---|
| RAM — what it really needs | 16 GB system RAM, and VRAM is what actually decides your model |
| CPU | Any modern CPU; the GPU is what matters |
| Disk | 50 GB+ — models are several gigabytes each |
| Monthly cost | $0 — it runs on the machine you already own |
| Setup time | 20 minutes |
| How you install it | Desktop installer; download models from the built-in catalogue |
| Ongoing maintenance | None beyond updates. |
| Where it stops scaling | One machine, one user. It is a desktop application by design. |
The thing that catches people out
The rule that decides everything is whether the quantised model fits in VRAM. A 7B model at Q4 needs roughly 4–5 GB and runs well; the same model at Q8 needs about 8 GB and will spill into system RAM on a smaller card, at which point throughput collapses by an order of magnitude with no error. Check the quantisation against your card before blaming the model.
When not to self-host LM Studio
You need an API for other services to call. LM Studio has a server mode, but Ollama is the better-supported path for that job.
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 QdrantVectors × dimensions × 4 bytes, in RAM, plus overhead — 1M × 768d is roughly 3 GB
- 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
Common questions
- How much RAM does LM Studio actually need?
- 16 GB system RAM, and VRAM is what actually decides your model in practice. The documented minimum is 8 GB, which is the figure at which the process starts rather than the figure at which it works under real use. Any modern CPU; the GPU is what matters alongside it.
- What does self-hosting LM Studio cost per month?
- $0 — it runs on the machine you already own This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for none beyond updates.
- How long does it take to set up LM Studio?
- 20 minutes, via Desktop installer; download models from the built-in catalogue.
- When should I NOT self-host LM Studio?
- You need an API for other services to call. LM Studio has a server mode, but Ollama is the better-supported path for that job.
- What is the most common mistake when self-hosting LM Studio?
- The rule that decides everything is whether the quantised model fits in VRAM. A 7B model at Q4 needs roughly 4–5 GB and runs well; the same model at Q8 needs about 8 GB and will spill into system RAM on a smaller card, at which point throughput collapses by an order of magnitude with no error. Check the quantisation against your card before blaming the model.