Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
157.4 tok/s
TTFT
1230 ms
Safe context
63K
Memory
32.4 GB / 128.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 170.5 tok/s | 1652 ms | 113K |
| Chat | C | Runs well | 170.5 tok/s | 619 ms | 34K |
| Coding | C | Runs well | 157.4 tok/s | 1230 ms | 63K |
| RAG | C | Runs well | 170.5 tok/s | 2065 ms | 113K |
| Reasoning | C | Runs well | 170.5 tok/s | 1342 ms | 63K |
How Devstral Small 2 24B Instruct (24B params) fits at each quantization level on AMD Instinct MI250X 128GB (128.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 9.4 GB | Low | D31 |
Q3_K_S | 3 | 11.8 GB | Low | D31 |
NVFP4 | 4 |
ollama run devstral-small-2-24bhuggingface-cli download devstral-small-2-24b13.4 GB |
| Medium |
| D32 |
Q4_K_M | 4 | 14.6 GB | Medium | D32 |
Q5_K_M | 5 | 17.3 GB | High | D32 |
Q6_K | 6 | 19.7 GB | High | D33 |
Q8_0 | 8 | 25.7 GB | Very High | D34 |
F16Best for your GPU | 16 | 49.2 GB | Maximum | D38 |