Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
89.5 tok/s
TTFT
2164 ms
Safe context
24K
Memory
21.3 GB / 32.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | B | Runs well | 89.5 tok/s | 3148 ms | 33K |
| Chat | C | Runs well | 89.5 tok/s | 1180 ms | 13K |
| Coding | B | Runs well | 89.5 tok/s | 2164 ms | 24K |
| RAG | B | Runs well | 89.5 tok/s | 3934 ms | 33K |
| Reasoning | B | Runs well | 89.5 tok/s | 2557 ms | 24K |
How Codestral 22B (22B params) fits at each quantization level on RTX 5090 32GB (32.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.6 GB | Low | D35 |
Q3_K_S | 3 | 10.8 GB | Low | D36 |
NVFP4 | 4 | 12.3 GB | Medium | D37 |
Q4_K_M | 4 | 13.4 GB | Medium | D38 |
Q5_K_M | 5 | 15.8 GB | High | D40 |
Q6_K | 6 | 18.0 GB | High | C41 |
Q8_0Best for your GPU | 8 | 23.5 GB | Very High | C44 |
F16 | 16 | 45.1 GB | Maximum | F0 |
ollama run codestral-22bhuggingface-cli download codestral-22b