Can it run?
Too heavy
Using Q4_K_M in vLLM
Fit status
Too heavy
Decode
5.5 tok/s
TTFT
35300 ms
Safe context
4K
Memory
425.4 GB / 48.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 5.5 tok/s | 51345 ms | 4K |
| Chat | F | Too heavy | 5.5 tok/s | 19254 ms | 4K |
| Coding | F | Too heavy | 5.5 tok/s | 35300 ms | 4K |
| RAG | F | Too heavy | 5.5 tok/s | 64181 ms | 4K |
| Reasoning | F | Too heavy | 5.5 tok/s | 41718 ms | 4K |
How Mistral Large 3 (675B params) fits at each quantization level on RTX 6000 Ada 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 263.3 GB | Low | F0 |
Q3_K_S | 3 | 330.8 GB | Low | F0 |
NVFP4 | 4 | 378.0 GB | Medium | F0 |
Q4_K_M | 4 | 411.8 GB | Medium | F0 |
Q5_K_M | 5 | 486.0 GB | High | F0 |
Q6_K | 6 | 553.5 GB | High | F0 |
Q8_0 | 8 | 722.3 GB | Very High | F0 |
F16 | 16 | 1383.7 GB | Maximum | F0 |
