Mac Studio M2 Ultra 128GBBest value
D6.7 tok/s decode
~$3,999 MSRP
Can it run?
Very compromised (needs ~11 GB host RAM)
Using Q4_K_M in vLLM
Fit status
Very compromised (needs ~11 GB host RAM)
Decode
33.0 tok/s
TTFT
5870 ms
Safe context
13K
Memory
95.5 GB / 80.0 GB
Offload
20%
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 35.9 tok/s | 7847 ms | 23K |
| Chat | C | Very compromised (needs ~5.3 GB host RAM) | 34.5 tok/s | 3062 ms | 7K |
| Coding | D | Very compromised (needs ~11 GB host RAM) | 33.0 tok/s | 5870 ms | 13K |
| RAG | F | Too heavy | 35.9 tok/s | 9808 ms | 23K |
| Reasoning | D | Very compromised (needs ~11 GB host RAM) | 33.0 tok/s | 6937 ms | 13K |
How Command A 111B (111B params) fits at each quantization level on NVIDIA H800 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 43.3 GB | Low | C41 |
Q3_K_SBest for your GPU | 3 | 54.4 GB | Low | C44 |
NVFP4 | 4 | 62.2 GB | Medium | C44 |
Q4_K_M | 4 | 67.7 GB | Medium | C44 |
Q5_K_M | 5 | 79.9 GB | High | C44 |
Q6_K | 6 | 91.0 GB | High | F0 |
Q8_0 | 8 | 118.8 GB | Very High | F0 |
F16 | 16 | 227.6 GB | Maximum | F0 |
ollama run command-a-111bhuggingface-cli download command-a-111bUpgrade options
~$3,999 MSRP
~$8,000 MSRP
NVIDIA H200 141GBBiggest leap~$30,000 MSRP
NVIDIA H200 PCIe 141GBNVIDIA upgrade~$30,000 MSRP