Can I Run Mistral Family on Beelink SER9 (Ryzen AI 9, 32 GB)?

Superseded model. Mistral Family has been superseded by Mistral Small 4. This page is kept for reference; the newer family is a better starting point. View Mistral Small 4 →

Written by Jakub Rusinowski · Last updated January 28, 2025

Yes

Yes — Mistral Small 3 (24B) at Q8_0 needs about 28.2 GB of the 32 GB usable on Beelink SER9 (Ryzen AI 9, 32 GB) (~3.8 GB spare), at ~7.1 tok/s (estimated), with room for about 16,384 tokens of context.

Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~7.1 tok/s

Beelink SER9 (Ryzen AI 9, 32 GB) — what it gives a model

Usable memory for models	32 GB
Memory bandwidth	256 GB/s
Form factor	Mini PC
Operating system	Windows or Linux
Memory upgradeable	Yes
Price	$859 (lib/data/ai-stations.ts (street price), checked 2026-07-06)

Mistral Family on Beelink SER9 (Ryzen AI 9, 32 GB): memory by quantization

Quant	Memory needed	Fits 32 GB?	Max context	Est. speed	Download
F16	50.7 GB	✗ No	—	—	48 GB
Q8_0	28.2 GB	✓ Yes	16K	~7.1 tok/s	25.5 GB
Q6_K	22.4 GB	✓ Yes	32K	~9.1 tok/s	19.7 GB
Q5_K_M	19.7 GB	✓ Yes	32K	~10.4 tok/s	17 GB
Q4_K_M	17.2 GB	✓ Yes	32K	~12.1 tok/s	14.5 GB
Q3_K_M	12.9 GB	✓ Yes	32K	~16.6 tok/s	10.2 GB
Q2_K	10.6 GB	✓ Yes	32K	~20.7 tok/s	7.9 GB

Which Mistral Family sizes fit

Variant	Needs at Q4_K_M	Fits?	Est. speed
Mistral Small 3 (24B)	17.2 GB	✓ Fits	~12.1 tok/s
Mistral NeMo 12B	9.4 GB	✓ Fits	~23 tok/s

Beelink SER9 32 GB limitations

Shares system memory with the iGPU, so the usable model budget is well below the nominal 32 GB.
Memory bandwidth, not capacity, is the limit here — expect single-digit tokens/sec on large models.

Recommended setup

Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving

How these numbers are calculated

Memory need = quantized weights + KV cache at 8,192 tokens (f16) + 0.8 GB runtime overhead.
32 GB unified memory at 256 GB/s, shared between CPU and GPU.
Throughput is a memory-bandwidth roofline estimate, not a measurement. It is labelled "estimated" everywhere it appears.
KV cache is computed from this model's published attention configuration.

FAQ