Written by Jakub Rusinowski · Last updated December 18, 2024
Yes — comfortably
Yes, comfortably — Falcon 3 10B Instruct at Q8_0 needs about 13.1 GB of the 24 GB usable on RTX 4090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~10.9 GB spare and running at ~56.8 tok/s (estimated), with room for about 32,768 tokens of context.
Confidence: high · Recommended quantization: Q8_0 · Estimated speed: ~56.8 tok/s
| Usable memory for models | 24 GB |
| Memory bandwidth | 1008 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 | 22.7 GB | ✓ Yes | 8K | ~32.9 tok/s | 20.6 GB |
| Q8_0 | 13.1 GB | ✓ Yes | 32K | ~56.8 tok/s | 10.9 GB |
| Q6_K | 10.6 GB | ✓ Yes | 32K | ~69.9 tok/s | 8.4 GB |
| Q5_K_M | 9.4 GB | ✓ Yes | 32K | ~78.2 tok/s | 7.3 GB |
| Q4_K_M | 8.4 GB | ✓ Yes | 32K | ~88 tok/s | 6.2 GB |
| Q3_K_M | 6.5 GB | ✓ Yes | 32K | ~111.8 tok/s | 4.4 GB |
| Q2_K | 5.5 GB | ✓ Yes | 32K | ~131.4 tok/s | 3.4 GB |
| Variant | Needs at Q4_K_M | Fits? | Est. speed |
|---|---|---|---|
| Falcon 3 10B Instruct | 8.4 GB | ✓ Fits | ~88 tok/s |
| Falcon 3 7B Instruct | 6.2 GB | ✓ Fits | ~113.3 tok/s |
| Falcon 3 3B Instruct | 3.5 GB | ✓ Fits | ~188.2 tok/s |
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
Yes, comfortably — Falcon 3 10B Instruct at Q8_0 needs about 13.1 GB of the 24 GB usable on RTX 4090 Desktop (24 GB VRAM, 64 GB RAM), leaving ~10.9 GB spare and running at ~56.8 tok/s (estimated), with room for about 32,768 tokens of context.
Q8_0 — it needs about 13.1 GB of the 24 GB available, downloads as roughly 10.9 GB, and runs at an estimated 56.8 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