Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
404.7 tok/s
TTFT
478 ms
Safe context
43K
Memory
29.5 GB / 80.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 404.7 tok/s | 696 ms | 86K |
| Chat | C | Runs well | 404.7 tok/s | 350 ms | 22K |
| Coding | C | Runs well | 404.7 tok/s | 478 ms | 43K |
| RAG | C | Runs well | 404.7 tok/s | 870 ms | 86K |
| Reasoning | C | Runs well | 404.7 tok/s | 565 ms | 43K |
How Qwen3-VL 30B A3B Instruct (30B params) fits at each quantization level on NVIDIA H100 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 11.7 GB | Low | D33 |
Q3_K_S | 3 | 14.7 GB | Low | D33 |
NVFP4 | 4 | 16.8 GB | Medium | D34 |
Q4_K_M | 4 | 18.3 GB | Medium | D34 |
Q5_K_M | 5 | 21.6 GB | High | D35 |
Q6_K | 6 | 24.6 GB | High | D36 |
Q8_0 | 8 | 32.1 GB | Very High | D38 |
F16Best for your GPU | 16 | 61.5 GB | Maximum | C45 |
huggingface-cli download qwen-3-vl-30b-a3b