Can it run?
Very compromised
Using Q4_K_M in Ollama
Fit status
Very compromised
Decode
21.0 tok/s
TTFT
9607 ms
Safe context
14K
Memory
334.2 GB / 288.0 GB
Offload
10%
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 24.1 tok/s | 11677 ms | 23K |
| Chat | D | Runs with offload (needs ~12.1 GB host RAM) | 23.1 tok/s | 4574 ms | 8K |
| Coding | D | Very compromised | 21.0 tok/s | 9607 ms | 14K |
| RAG | F | Too heavy | 24.1 tok/s | 14597 ms | 23K |
| Reasoning | D | Very compromised (needs ~33.5 GB host RAM) | 21.3 tok/s | 10751 ms |
How Qwen3.5 397B A17B (397B params) fits at each quantization level on AMD Instinct MI350X 288GB (288.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 154.8 GB | Low | C41 |
Q3_K_S | 3 | 194.5 GB | Low | C44 |
NVFP4Best for your GPU |
| 14K |
| 4 |
222.3 GB |
| Medium |
| C44 |
Q4_K_M | 4 | 242.2 GB | Medium | C44 |
Q5_K_M | 5 | 285.8 GB | High | C44 |
Q6_K | 6 | 325.5 GB | High | F0 |
Q8_0 | 8 | 424.8 GB | Very High | F0 |
F16 | 16 | 813.8 GB | Maximum | F0 |