NVIDIA H100 80GBBudget pickC99.7 tok/s decode
~$40,000 MSRP
Can it run?
Runs well
Using Q5_K_M in Ollama
Fit status
Runs well
Decode
39.4 tok/s
TTFT
4908 ms
Safe context
8K
Memory
42.7 GB / 64.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 39.4 tok/s | 7138 ms | 8K |
| Chat | C | Runs well | 39.4 tok/s | 2677 ms | 8K |
| Coding | C | Runs well | 39.4 tok/s | 4908 ms | 8K |
| RAG | C | Runs well | 39.4 tok/s | 8923 ms | 8K |
| Reasoning | C | Runs well | 39.4 tok/s | 5800 ms | 8K |
How Falcon 40B Instruct (40B params) fits at each quantization level on AMD Instinct MI210 64GB (64.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 15.6 GB | Low | D34 |
Q3_K_S | 3 | 19.6 GB | Low | D35 |
NVFP4 | 4 | 22.4 GB | Medium | D36 |
Q4_K_M | 4 | 24.4 GB | Medium | D37 |
Q5_K_M | 5 | 28.8 GB | High | D38 |
Q6_K | 6 | 32.8 GB | High | D40 |
Q8_0Best for your GPU | 8 | 42.8 GB | Very High | C43 |
F16 | 16 | 82.0 GB | Maximum | F0 |
ollama run falcon-40b-instructhuggingface-cli download falcon-40b-instructUpgrade options
NVIDIA H100 80GBBudget pick~$40,000 MSRP
NVIDIA H800 80GBBiggest leap