Mac Studio M2 Ultra 128GBBudget pick
C23.8 tok/s decode
~$3,999 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
11.9 tok/s
TTFT
16289 ms
Safe context
31K
Memory
36.1 GB / 69.1 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 11.9 tok/s | 23693 ms | 54K |
| Chat | C | Runs well | 11.9 tok/s | 8885 ms | 16K |
| Coding | C | Runs well | 11.9 tok/s | 16289 ms | 31K |
| RAG | C | Runs well | 11.9 tok/s | 29617 ms | 54K |
| Reasoning | C | Runs well | 11.9 tok/s | 19251 ms | 31K |
How Qwen 2.5 Coder 32B (32B params) fits at each quantization level on MacBook Pro M2 Max 96GB (69.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 12.5 GB | Low | D33 |
Q3_K_S | 3 | 15.7 GB | Low | D34 |
NVFP4 | 4 | 17.9 GB | Medium | D35 |
Q4_K_M | 4 | 19.5 GB | Medium | D35 |
Q5_K_M | 5 | 23.0 GB | High | D37 |
Q6_K | 6 | 26.2 GB | High | D38 |
Q8_0Best for your GPU | 8 | 34.2 GB | Very High | C40 |
F16 | 16 | 65.6 GB | Maximum | C44 |
ollama run qwen-2.5-coder-32bhuggingface-cli download qwen-2.5-coder-32bUpgrade options
~$3,999 MSRP
NVIDIA A100 80GBBest value~$15,000 MSRP
NVIDIA H100 80GBBiggest leap~$40,000 MSRP