Calculate Whether a Local AI Model Fits in Memory
12 models by download size, 32 NAS models by memory
As an Amazon Associate we earn from qualifying purchases. Prices and availability come from Amazon.com as of the last update and may change.
Pick a model and a NAS. The calculator takes the model's download size, as Ollama's library prints it, adds the memory you choose to keep back, and subtracts both from the memory in the NAS. The download size is the starting point because Ollama's FAQ says a model shown as 100% CPU was loaded entirely in system memory. With Llama 3.1 8B at 4.9GB and 2GB kept back, 16 of the 32 NAS models on record have enough memory as shipped.
What goes into the sum
| Input | Where it comes from |
|---|---|
| Model download size | Ollama's library page for that tag |
| Memory in the NAS | The maker's figure for the model as shipped; type your own if you have added memory |
| Memory you keep back | Yours to set, for the NAS's own system, its other apps and the model's context window |
The models in the list
| Model | Ollama tag | Download |
|---|---|---|
| Llama 3.2 1B | llama3.2:1b | 1.3GB |
| Llama 3.2 3B | llama3.2:3b | 2.0GB |
| Gemma 3 4B | gemma3:4b | 3.4GB |
| Mistral 7B | mistral:7b | 4.4GB |
| Llama 3.1 8B | llama3.1:8b | 4.9GB |
| Qwen3 8B | qwen3:8b | 5.2GB |
| Gemma 3 12B | gemma3:12b | 8.2GB |
| Phi-4 14B | phi4:14b | 9.1GB |
| Qwen3 14B | qwen3:14b | 9.3GB |
| Gemma 3 27B | gemma3:27b | 17GB |
| DeepSeek-R1 32B | deepseek-r1:32b | 20GB |
| Llama 3.1 70B | llama3.1:70b | 43GB |
What the result does and does not tell you
It tells you whether the file fits beside whatever you keep back. Ollama states that required memory scales with the context window and the number of parallel requests, and the pages read here give no figure for either, so the calculator does not invent one. Raise the amount kept back if you raise the context window from its 4,096-token default.
It says nothing about speed. No source used on this page states how fast any model runs on any NAS.
Related pages
The same sizes at other quantization levels, and the Ollama settings behind the sum.
Where the figures come from
Ollama's FAQ and library tag pages, read on October 12, 2026. NAS memory figures are each maker's, as recorded on this site's device pages.
Checked on October 12, 2026. Sources: Ollama documentation: FAQ; Ollama library: llama3.2 tags; Ollama library: llama3.1 tags; Ollama library: gemma3 tags; Ollama library: qwen3 tags; Ollama library: deepseek-r1 tags; Ollama library: mistral tags; Ollama library: phi4 tags.
Of the 32 NAS models on record, 26 ship with at least 3.3GB and 5 more reach it at their maker's memory limit.
The amount you keep back is your own allowance for the NAS's system, its other apps and the model's context window. Ollama gives no figure for any of them on a NAS, so set it from what your own unit reports as in use.
NAS on Amazon.com, by price
Every new NAS listing tracked, cheapest first. Memory figures on this page are for the models named in its tables only.
Frequently asked questions
Is the download size the same as the memory a model uses?
It is the starting point, not the whole figure. Ollama says a model on the CPU is loaded entirely in system memory and that required memory also scales with the context window.
What should I enter for memory kept back?
What your own NAS shows as in use before the model loads, plus an allowance for the context window. Ollama publishes no figure for either on a NAS.
Does this calculator cover graphics card memory?
No. It compares a download with system memory, which is where Ollama loads a model when no graphics card is used.