Written by Jakub Rusinowski · Last updated October 16, 2024
Yes — comfortably
Yes, comfortably — Ministral 8B at Q8_0 needs about 10.5 GB of the 24 GB usable on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~13.5 GB spare and running at ~65.6 tok/s (estimated), with room for about 32,768 tokens of context.
Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~65.6 tok/s
| Usable memory for models | 24 GB |
| Memory bandwidth | 936 GB/s |
| Form factor | Desktop |
| Operating system | Windows or Linux |
| Memory upgradeable | Yes |
| Quant | Memory needed | Fits 24 GB? | Max context | Est. speed | Download |
|---|---|---|---|---|---|
| F16 | 18 GB | ✓ Yes | 32K | ~38.5 tok/s | 16 GB |
| Q8_0 | 10.5 GB | ✓ Yes | 32K | ~65.6 tok/s | 8.5 GB |
| Q6_K | 8.6 GB | ✓ Yes | 32K | ~80.2 tok/s | 6.6 GB |
| Q5_K_M | 7.7 GB | ✓ Yes | 32K | ~89.3 tok/s | 5.7 GB |
| Q4_K_M | 6.9 GB | ✓ Yes | 32K | ~100 tok/s | 4.8 GB |
| Q3_K_M | 5.4 GB | ✓ Yes | 32K | ~125.4 tok/s | 3.4 GB |
| Q2_K | 4.6 GB | ✓ Yes | 32K | ~145.7 tok/s | 2.6 GB |
| Variant | Needs at Q4_K_M | Fits? | Est. speed |
|---|---|---|---|
| Ministral 8B | 6.9 GB | ✓ Fits | ~100 tok/s |
| Ministral 3B | 3.8 GB | ✓ Fits | ~167.7 tok/s |
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
Yes, comfortably — Ministral 8B at Q8_0 needs about 10.5 GB of the 24 GB usable on RTX 3090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~13.5 GB spare and running at ~65.6 tok/s (estimated), with room for about 32,768 tokens of context.
Q8_0 — it needs about 10.5 GB of the 24 GB available, downloads as roughly 8.5 GB, and runs at an estimated 65.6 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? | Ministral model page | Check your hardware