Written by Jakub Rusinowski · Last updated March 17, 2025
Yes
Yes — Mistral Small 3.1 24B at Q8_0 needs about 27.2 GB of the 32 GB usable on RTX 5090 Desktop (32 GB VRAM, 64 GB RAM) (~4.8 GB spare), at ~46.7 tok/s (estimated), with room for about 32,768 tokens of context.
Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~46.7 tok/s
| Usable memory for models | 32 GB |
| Memory bandwidth | 1792 GB/s |
| Form factor | Desktop |
| Operating system | Windows or Linux |
| Memory upgradeable | Yes |
| Quant | Memory needed | Fits 32 GB? | Max context | Est. speed | Download |
|---|---|---|---|---|---|
| F16 | 49.3 GB | ✗ No | — | — | 47.2 GB |
| Q8_0 | 27.2 GB | ✓ Yes | 32K | ~46.7 tok/s | 25.1 GB |
| Q6_K | 21.5 GB | ✓ Yes | 64K | ~58.3 tok/s | 19.4 GB |
| Q5_K_M | 18.9 GB | ✓ Yes | 64K | ~65.8 tok/s | 16.7 GB |
| Q4_K_M | 16.4 GB | ✓ Yes | 64K | ~74.9 tok/s | 14.2 GB |
| Q3_K_M | 12.2 GB | ✓ Yes | 64K | ~97.8 tok/s | 10.1 GB |
| Q2_K | 9.9 GB | ✓ Yes | 64K | ~117.4 tok/s | 7.8 GB |
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
Yes — Mistral Small 3.1 24B at Q8_0 needs about 27.2 GB of the 32 GB usable on RTX 5090 Desktop (32 GB VRAM, 64 GB RAM) (~4.8 GB spare), at ~46.7 tok/s (estimated), with room for about 32,768 tokens of context.
Q8_0 — it needs about 27.2 GB of the 32 GB available, downloads as roughly 25.1 GB, and runs at an estimated 46.7 tokens/sec with up to 32K of context.
Nothing binding — the model fits with headroom and generates at a usable speed on this hardware.
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
← Can I Run It? | Mistral Small 3.1 model page | Check your hardware