AMD Instinct MI350X 288GBBest value
D21.3 tok/s decode
~$8,000 MSRP
Can it run?
Too heavy
Using Q4_K_M in Ollama
Fit status
Too heavy
Decode
2.6 tok/s
TTFT
75367 ms
Safe context
4K
Memory
310.2 GB / 48.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 2.8 tok/s | 101192 ms | 4K |
| Chat | F | Too heavy | 2.8 tok/s | 37947 ms | 4K |
| Coding | F | Too heavy | 2.6 tok/s | 75367 ms | 4K |
| RAG | F | Too heavy | 2.8 tok/s | 126490 ms | 4K |
| Reasoning | F | Too heavy | 2.8 tok/s | 82218 ms | 4K |
How Qwen3.5 397B A17B (397B params) fits at each quantization level on NVIDIA L40 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 154.8 GB | Low | F0 |
Q3_K_S | 3 | 194.5 GB | Low | F0 |
NVFP4 | 4 |
Upgrade options
~$8,000 MSRP
| Medium |
| F0 |
Q4_K_M | 4 | 242.2 GB | Medium | F0 |
Q5_K_M | 5 | 285.8 GB | High | F0 |
Q6_K | 6 | 325.5 GB | High | F0 |
Q8_0 | 8 | 424.8 GB | Very High | F0 |
F16 | 16 | 813.8 GB | Maximum | F0 |