Layer 4 · self-hosting reality check
What it actually takes to self-host ClearML
The docs say 4 GB. In practice you want 8 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 Weights & Biases alternatives, where ClearML is one of the picks.
| RAM — documented minimum | 4 GB |
|---|---|
| RAM — what it really needs | 8 GB |
| CPU | 4 vCPU |
| Disk | Artifacts plus MongoDB and Elasticsearch storage |
| Monthly cost | $50–80/mo self-hosted, against per-seat MLOps platform pricing |
| Setup time | Half a day |
| How you install it | docker compose — app server, file server, MongoDB, Elasticsearch, Redis |
| Ongoing maintenance | Moderate to high. Elasticsearch is the component that will need attention. |
| Where it stops scaling | Full MLOps at team scale. The broadest scope of anything in this batch and correspondingly the most to operate. |
The thing that catches people out
Its automatic logging captures almost everything with no instrumentation, which is the selling point — and it means it will also capture credentials passed as command-line arguments or environment variables straight into the experiment record, visible to everyone with access. Configure the hidden-argument list before pointing it at anything real.
When not to self-host ClearML
You only need experiment tracking. ClearML's breadth — orchestration, data versioning, serving — is the reason it is heavy, and MLflow or Aim are lighter if you will not use the rest.
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 ClearML actually need?
- 8 GB in practice. The documented minimum is 4 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 ClearML cost per month?
- $50–80/mo self-hosted, against per-seat MLOps platform pricing This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for moderate to high. Elasticsearch is the component that will need attention.
- How long does it take to set up ClearML?
- Half a day, via docker compose — app server, file server, MongoDB, Elasticsearch, Redis.
- When should I NOT self-host ClearML?
- You only need experiment tracking. ClearML's breadth — orchestration, data versioning, serving — is the reason it is heavy, and MLflow or Aim are lighter if you will not use the rest.
- What is the most common mistake when self-hosting ClearML?
- Its automatic logging captures almost everything with no instrumentation, which is the selling point — and it means it will also capture credentials passed as command-line arguments or environment variables straight into the experiment record, visible to everyone with access. Configure the hidden-argument list before pointing it at anything real.