Can it run?
Runs well
Using Q5_K_M in Ollama
Fit status
Runs well
Decode
32.6 tok/s
TTFT
5931 ms
Safe context
8K
Memory
32.3 GB / 48.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 37.2 tok/s | 7575 ms | 8K |
| Chat | C | Runs well | 37.2 tok/s | 2841 ms | 8K |
| Coding | C | Runs well | 32.6 tok/s | 5931 ms | 8K |
| RAG | C | Runs well | 37.2 tok/s | 9469 ms | 8K |
| Reasoning | C | Runs well | 37.2 tok/s | 6155 ms | 8K |
How MPT-30B-Instruct (30B params) fits at each quantization level on RTX 6000 Ada 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 11.7 GB | Low | D34 |
Q3_K_S | 3 | 14.7 GB | Low | D35 |
NVFP4 | 4 |
ollama run mpt-30b-instructhuggingface-cli download mpt-30b-instruct16.8 GB |
| Medium |
| D36 |
Q4_K_M | 4 | 18.3 GB | Medium | D37 |
Q5_K_M | 5 | 21.6 GB | High | D38 |
Q6_K | 6 | 24.6 GB | High | D40 |
Q8_0Best for your GPU | 8 | 32.1 GB | Very High | C43 |
F16 | 16 | 61.5 GB | Maximum | F0 |