Can it run?
Runs well
Using Q5_K_M in Ollama
Fit status
Runs well
Decode
99.7 tok/s
TTFT
1943 ms
Safe context
8K
Memory
44.3 GB / 80.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 99.7 tok/s | 2826 ms | 8K |
| Chat | C | Runs well | 99.7 tok/s | 1060 ms | 8K |
| Coding | C | Runs well | 99.7 tok/s | 1943 ms | 8K |
| RAG | C | Runs well | 99.7 tok/s | 3532 ms | 8K |
| Reasoning | C | Runs well | 99.7 tok/s | 2296 ms | 8K |
How Falcon 40B Instruct (40B params) fits at each quantization level on NVIDIA H100 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 15.6 GB | Low | D33 |
Q3_K_S | 3 | 19.6 GB | Low | D34 |
NVFP4 | 4 | 22.4 GB | Medium | D35 |
Q4_K_M | 4 | 24.4 GB | Medium | D35 |
Q5_K_M | 5 | 28.8 GB | High | D36 |
Q6_K | 6 | 32.8 GB | High | D37 |
Q8_0Best for your GPU | 8 | 42.8 GB | Very High | C40 |
F16 | 16 | 82.0 GB | Maximum | F0 |
ollama run falcon-40b-instructhuggingface-cli download falcon-40b-instruct