MacBook Pro M3 Max 128GBBest value
C3.7 tok/s decode
~$2,499 MSRP
Can it run?
Very compromised (needs ~5.5 GB host RAM)
Using Q3_K_M in Ollama
Fit status
Very compromised (needs ~5.5 GB host RAM)
Decode
36.1 tok/s
TTFT
5366 ms
Safe context
15K
Memory
88.0 GB / 80.0 GB
Offload
10%
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 37.8 tok/s | 7449 ms | 24K |
| Chat | C | Runs with offload | 37.8 tok/s | 2793 ms | 8K |
| Coding | C | Very compromised (needs ~5.5 GB host RAM) | 36.1 tok/s | 5366 ms | 15K |
| RAG | F | Too heavy | 37.8 tok/s | 9311 ms | 24K |
| Reasoning | C | Very compromised (needs ~5.5 GB host RAM) | 36.1 tok/s | 6342 ms | 15K |
How Qwen3.5 122B A10B (122B params) fits at each quantization level on NVIDIA H800 80GB (80.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 47.6 GB | Low | C42 |
Q3_K_SBest for your GPU | 3 | 59.8 GB | Low | C45 |
NVFP4 | 4 | 68.3 GB | Medium | C45 |
Q4_K_M | 4 | 74.4 GB | Medium | C45 |
Q5_K_M | 5 | 87.8 GB | High | F0 |
Q6_K | 6 | 100.0 GB | High | F0 |
Q8_0 | 8 | 130.5 GB | Very High | F0 |
F16 | 16 | 250.1 GB | Maximum | F0 |
huggingface-cli download hf-unsloth--qwen3-5-122b-a10b-ggufUpgrade options
~$2,499 MSRP
~$8,000 MSRP
NVIDIA H200 141GBNVIDIA upgrade~$30,000 MSRP