NVIDIA GH200 96GBBudget pickB73.8 tok/s decode
Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
39.0 tok/s
TTFT
4964 ms
Safe context
20K
Memory
65.6 GB / 80.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs with offload | 39.0 tok/s | 7221 ms | 33K |
| Chat | C | Runs well | 39.0 tok/s | 2708 ms | 11K |
| Coding | C | Runs well | 39.0 tok/s | 4964 ms | 20K |
| RAG | C | Runs with offload | 39.0 tok/s | 9026 ms | 33K |
| Reasoning | C | Runs well | 39.0 tok/s | 5867 ms | 20K |
How Qwen 2.5 VL 72B (72B params) fits at each quantization level on NVIDIA A100 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 28.1 GB | Low | D37 |
Q3_K_S | 3 | 35.3 GB | Low | D39 |
NVFP4 | 4 | 40.3 GB | Medium | C40 |
Q4_K_M | 4 | 43.9 GB | Medium | C41 |
Q5_K_M | 5 | 51.8 GB | High | C43 |
Q6_KBest for your GPU | 6 | 59.0 GB | High | C44 |
Q8_0 | 8 | 77.0 GB | Very High | C44 |
F16 | 16 | 147.6 GB | Maximum | F0 |
ollama run qwen-2.5-vl-72bhuggingface-cli download qwen-2.5-vl-72bUpgrade options
NVIDIA GH200 96GBBudget pick
NVIDIA H20 96GBBiggest leap