NVIDIA H100 80GBBudget pickC96.1 tok/s decode
~$40,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
38.0 tok/s
TTFT
5089 ms
Safe context
23K
Memory
44.4 GB / 64.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 38.0 tok/s | 7403 ms | 39K |
| Chat | C | Runs well | 38.0 tok/s | 2776 ms | 13K |
| Coding | C | Runs well | 38.0 tok/s | 5089 ms | 23K |
| RAG | C | Runs well | 38.0 tok/s | 9253 ms | 39K |
| Reasoning | C | Runs well | 38.0 tok/s | 6015 ms | 23K |
How Qwen3 48B A4B Savant Commander Distill 12X Closed Open Heretic Uncensored (48B params) fits at each quantization level on AMD Instinct MI210 64GB (64.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 18.7 GB | Low | D35 |
Q3_K_S | 3 | 23.5 GB | Low | D37 |
NVFP4 | 4 | 26.9 GB | Medium | D38 |
Q4_K_M | 4 | 29.3 GB | Medium | D39 |
Q5_K_M | 5 | 34.6 GB | High | C41 |
Q6_K | 6 | 39.4 GB | High | C42 |
Q8_0Best for your GPU | 8 | 51.4 GB | Very High | C44 |
F16 | 16 | 98.4 GB | Maximum | F0 |
Upgrade options
NVIDIA H100 80GBBudget pick~$40,000 MSRP
NVIDIA H800 80GBBiggest leap