Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
299.2 tok/s
TTFT
647 ms
Safe context
91K
Memory
45.2 GB / 256.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 299.2 tok/s | 941 ms | 167K |
| Chat | C | Runs well | 299.2 tok/s | 353 ms | 47K |
| Coding | C | Runs well | 299.2 tok/s | 647 ms | 91K |
| RAG | C | Runs well | 299.2 tok/s | 1177 ms | 167K |
| Reasoning | C | Runs well | 299.2 tok/s | 765 ms | 91K |
How Devstral Small 2 24B Instruct (24B params) fits at each quantization level on AMD Instinct MI325X 256GB (256.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 9.4 GB | Low | D30 |
Q3_K_S | 3 | 11.8 GB | Low | D30 |
NVFP4 | 4 | 13.4 GB | Medium | D31 |
Q4_K_M | 4 | 14.6 GB | Medium | D31 |
Q5_K_M | 5 | 17.3 GB | High | D31 |
Q6_K | 6 | 19.7 GB | High | D31 |
Q8_0 | 8 | 25.7 GB | Very High | D32 |
F16Best for your GPU | 16 | 49.2 GB | Maximum | D34 |
ollama run devstral-small-2-24bhuggingface-cli download devstral-small-2-24b