How much RAM does my vector database need?
A vector index wants to live in RAM. The graph walk touches memory in a scattered pattern, so spilling to disk costs far more than the size ratio suggests.
The short answer
Computed from this tool’s default settings — vectors stored and the rest as most people start. Change them below for your own case.
1,000,000 vectors need about 4.29 GB, comfortably inside 25.6 GB. You have room for roughly 5,973,269 at this dimension.
Vector Database Memory Calculator
Whether the index fits in memory, and what happens when it does not.
1,000,000 vectors need about 4.29 GB, comfortably inside 25.6 GB. You have room for roughly 5,973,269 at this dimension.
What each setting changes
Every input moves the result for a reason. This is what each one does and where to find the value for your own machine.
| Setting | Default | What it changes |
|---|---|---|
| Vectors stored | 1000000 | Your own figure, starting from 1,000,000. Change it to match what you actually run. |
| Embedding dimension | 768 | 5 options, from 384 to 3072. |
| Bytes per value | float32 | 3 options, from float32 to int8. |
| Available RAM | 32 GB | Anywhere from 2 to 512 GB. |
| Metadata per vector | 200 bytes | Source text, ids and filters stored alongside each vector. |
Real answers across embedding dimension
The same calculation run at a range of settings, with everything else left at its default. These are computed by the tool itself, not written by hand.
| Embedding dimension | Fits in RAM | Vectors | Graph | Metadata |
|---|---|---|---|---|
| 384 | 2.24 GB | 1.54 GB | 0.51 GB | 0.20 GB |
| 768 | 4.29 GB | 3.07 GB | 1.01 GB | 0.20 GB |
| 1024 | 5.65 GB | 4.10 GB | 1.35 GB | 0.20 GB |
| 1536 | 8.37 GB | 6.14 GB | 2.03 GB | 0.20 GB |
| 3072 | 16.5 GB | 12.3 GB | 4.06 GB | 0.20 GB |
1,000,000 vectors need about 2.24 GB, comfortably inside 25.6 GB. You have room for roughly 11,413,896 at this dimension.
How this is calculated
No lookup tables and no invented constants. Here is the arithmetic, so you can check it against your own numbers.
HNSW-style indexes want to be resident in RAM: the graph walk touches memory in a scattered pattern, so spilling to disk costs far more than the size ratio suggests. The question is binary — does it fit.
Metadata is easy to forget and often exceeds the vectors themselves when the source text is stored inline. Keep the text in a separate store and hold only ids in the index if memory is tight.
These are well-founded engineering estimates, not benchmark results. Your quantisation, runtime and context length all move the real number, and usable memory is an assumption rather than a specification. See the full methodology for every assumption behind these figures.
Who this is for
The three situations that bring people to this calculation.
Sizing a server
Check the index fits before deploying.
Scaling up
Find the vector ceiling for your RAM.
Slow search
Confirm whether you have started paging.
How to use this calculator
Four steps, no account, nothing leaves your browser.
Set vectors stored
Start at the top of the panel. Every figure recalculates as you change it — there is no submit button, because watching the number move is the point.
Adjust the rest to match your setup
4 further settings: embedding dimension, bytes per value, available ram, metadata per vector. Defaults are the common case, so change only what differs for you.
Read the headline, then the table
The large figure answers the question. The table underneath shows how the answer changes across nearby settings, which is usually where the decision actually gets made.
Check it against the method
The arithmetic is written out above. If a number looks wrong for your hardware, the assumptions are the first place to look — usable memory and quantisation are the two that vary most.
How much RAM does my vector database need: common questions
The questions people ask about this, answered without hedging.
How much RAM does a vector database need?
The vectors, plus roughly a third again for the graph, plus metadata. A million 768-dimension float32 vectors is around 4 GB before the text you store with them.
What happens if the index does not fit?
Search slows disproportionately. Graph traversal jumps around memory, so page faults hit constantly rather than streaming — it is much worse than a linear slowdown.
How do I fit more vectors?
Quantise to int8, move stored text out of the index and keep only ids, or shard across machines. Quantisation is usually the cheapest first move.
Is metadata really a problem?
Often it is the largest component. Storing source text inline next to each vector can exceed the vectors themselves several times over.
Related local AI tools
All 50 run on the same arithmetic, so answers across them agree.
More on context
Will it fit
Speed
Quantisation
Runtime & errors
Cost
Fine-tuning
Searches this page answers
Different ways of asking the same question, all resolved above.
Now find the models that fit
Sizing is only half the problem. Model Radar takes your hardware and shows which models actually run on it, ranked by what they are good at — the same arithmetic as this page, applied to every model worth running.