Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
193.0 tok/s
TTFT
1003 ms
Safe context
64K
Memory
32.1 GB / 128.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 193.0 tok/s | 1459 ms | 115K |
| Chat | C | Runs well | 193.0 tok/s | 547 ms | 34K |
| Coding | C | Runs well | 193.0 tok/s | 1003 ms | 64K |
| RAG | C | Runs well | 193.0 tok/s | 1824 ms | 115K |
| Reasoning | C | Runs well | 193.0 tok/s | 1186 ms | 64K |
How Codestral 2 25.08 (22B params) fits at each quantization level on Gaudi 3 128GB (128.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.6 GB | Low | D31 |
Q3_K_S | 3 | 10.8 GB | Low | D31 |
NVFP4 | 4 |
huggingface-cli download codestral-2-25.08| Medium |
| D31 |
Q4_K_M | 4 | 13.4 GB | Medium | D32 |
Q5_K_M | 5 | 15.8 GB | High | D32 |
Q6_K | 6 | 18.0 GB | High | D32 |
Q8_0 | 8 | 23.5 GB | Very High | D33 |
F16Best for your GPU | 16 | 45.1 GB | Maximum | D37 |