Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
125.2 tok/s
TTFT
1546 ms
Safe context
47K
Memory
27.3 GB / 80.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 125.2 tok/s | 2249 ms | 83K |
| Chat | C | Runs well | 125.2 tok/s | 844 ms | 25K |
| Coding | C | Runs well | 125.2 tok/s | 1546 ms | 47K |
| RAG | C | Runs well | 125.2 tok/s | 2812 ms | 83K |
| Reasoning | C | Runs well | 125.2 tok/s | 1828 ms | 47K |
How Codestral 2 25.08 (22B params) fits at each quantization level on NVIDIA H100 PCIe 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.6 GB | Low | D32 |
Q3_K_S | 3 | 10.8 GB | Low | D32 |
NVFP4 | 4 | 12.3 GB | Medium | D33 |
Q4_K_M | 4 | 13.4 GB | Medium | D33 |
Q5_K_M | 5 | 15.8 GB | High | D34 |
Q6_K | 6 | 18.0 GB | High | D34 |
Q8_0 | 8 | 23.5 GB | Very High | D36 |
F16Best for your GPU | 16 | 45.1 GB | Maximum | C41 |
huggingface-cli download codestral-2-25.08