Written by Jakub Rusinowski · Last updated May 29, 2024
Yes
Yes — Codestral 22B at Q6_K needs about 20.9 GB of the 24 GB usable on RTX 4090 Desktop (24 GB VRAM, 64 GB RAM) (~3.1 GB spare), at ~36.3 tok/s (estimated), with room for about 16,384 tokens of context.
Confidence: high · Recommended quantization: Q6_K · Estimated speed: ~36.3 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 | 47.1 GB | ✗ No | — | — | 44.4 GB |
| Q8_0 | 26.3 GB | ✗ No | — | — | 23.6 GB |
| Q6_K | 20.9 GB | ✓ Yes | 16K | ~36.3 tok/s | 18.2 GB |
| Q5_K_M | 18.4 GB | ✓ Yes | 16K | ~41.2 tok/s | 15.7 GB |
| Q4_K_M | 16.1 GB | ✓ Yes | 32K | ~47.1 tok/s | 13.4 GB |
| Q3_K_M | 12.1 GB | ✓ Yes | 32K | ~62.5 tok/s | 9.5 GB |
| Q2_K | 10 GB | ✓ Yes | 32K | ~76.1 tok/s | 7.3 GB |
Ollama or llama.cpp (CUDA/ROCm) — vLLM if you need concurrent serving
Yes — Codestral 22B at Q6_K needs about 20.9 GB of the 24 GB usable on RTX 4090 Desktop (24 GB VRAM, 64 GB RAM) (~3.1 GB spare), at ~36.3 tok/s (estimated), with room for about 16,384 tokens of context.
Q6_K — it needs about 20.9 GB of the 24 GB available, downloads as roughly 18.2 GB, and runs at an estimated 36.3 tokens/sec with up to 16K 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? | Codestral model page | Check your hardware