AMD Instinct MI250X 128GBBudget pick
B100 tok/s decode
~$15,000 MSRP
Can it run?
Runs with offload (needs ~3.5 GB host RAM)
Using Q4_K_M in vLLM
Fit status
Runs with offload (needs ~3.5 GB host RAM)
Decode
64.8 tok/s
TTFT
2989 ms
Safe context
15K
Memory
84.0 GB / 80.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs with offload (needs ~4.3 GB host RAM) | 64.2 tok/s | 4389 ms | 30K |
| Chat | C | Runs with offload (needs ~3.3 GB host RAM) | 64.9 tok/s | 1627 ms | 8K |
| Coding | C | Runs with offload (needs ~3.5 GB host RAM) | 64.8 tok/s | 2989 ms | 15K |
| RAG | C | Runs with offload (needs ~4.3 GB host RAM) | 64.2 tok/s | 5487 ms | 30K |
| Reasoning | C | Runs with offload (needs ~3.5 GB host RAM) | 64.8 tok/s | 3533 ms | 15K |
How Mistral Small 4 119B (119B params) fits at each quantization level on NVIDIA H100 PCIe 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 46.4 GB | Low | C42 |
Q3_K_SBest for your GPU | 3 | 58.3 GB | Low | C45 |
NVFP4 | 4 | 66.6 GB | Medium | C45 |
Q4_K_M | 4 | 72.6 GB | Medium | C45 |
Q5_K_M | 5 | 85.7 GB | High | F0 |
Q6_K | 6 | 97.6 GB | High | F0 |
Q8_0 | 8 | 127.3 GB | Very High | F0 |
F16 | 16 | 244.0 GB | Maximum | F0 |
huggingface-cli download mistral-small-4-119bUpgrade options
~$15,000 MSRP
NVIDIA H200 141GBBest value~$30,000 MSRP
NVIDIA H200 PCIe 141GBNVIDIA upgrade~$30,000 MSRP
Gaudi 3 128GBBiggest leap