Running Qwen models on a NAS

Which Qwen model should you run on a NAS?

Qwen covers every NAS size. Qwen3 4B and 8B suit 16GB systems and run quickly on the CPU. Qwen3 14B is a strong step up for 32GB systems. Qwen3 32B, at about 20GB at 4-bit, is one of the best models that fits a 64GB NAS comfortably. Qwen2.5 72B needs a 96GB-class system.

Why is Qwen3 30B-A3B popular on NAS hardware?

It is a mixture-of-experts model with about 30B total and 3.3B active parameters. It needs memory like a 30B model, around 19GB at 4-bit, but only a small part works on each word, so it answers much faster on a CPU than a dense model of similar size. That suits NAS systems without a GPU well.

What about Qwen3 235B?

At around 142GB at 4-bit, it needs server-class memory such as QNAP's h1290FX systems (Estimated).

Can a NAS run Qwen without a GPU?

Yes. The 4B, 8B and 30B-A3B models are comfortable on modern AI NAS processors. Dense 14B and 32B models work but answer more slowly.

Common questions

Does the Finder understand Qwen?

Yes. Choose Qwen in the Finder's AI model field, or a specific size, and it only shows systems with enough memory.

Which Qwen version is newer?

Qwen3 is the newer generation. Qwen2.5 remains useful for its 72B model.