AMD Instinct MI250X 128GBBudget pick
C127.9 tok/s decode
~$15,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
23.8 tok/s
TTFT
8145 ms
Safe context
37K
Memory
39.5 GB / 92.2 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 23.8 tok/s | 11847 ms | 66K |
| Chat | C | Runs well | 23.8 tok/s | 4442 ms | 20K |
| Coding | C | Runs well | 23.8 tok/s | 8145 ms | 37K |
| RAG | C | Runs well | 23.8 tok/s | 14808 ms | 66K |
| Reasoning | C | Runs well | 23.8 tok/s | 9625 ms | 37K |
How Qwen 2.5 Coder 32B (32B params) fits at each quantization level on Mac Studio M2 Ultra 128GB (92.2 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 12.5 GB | Low | D32 |
Q3_K_S | 3 | 15.7 GB | Low | D33 |
NVFP4 | 4 | 17.9 GB | Medium | D34 |
Q4_K_M | 4 | 19.5 GB | Medium | D34 |
Q5_K_M | 5 | 23.0 GB | High | D35 |
Q6_K | 6 | 26.2 GB | High | D36 |
Q8_0 | 8 | 34.2 GB | Very High | D37 |
F16Best for your GPU | 16 | 65.6 GB | Maximum | C44 |
ollama run qwen-2.5-coder-32bhuggingface-cli download qwen-2.5-coder-32bUpgrade options
~$15,000 MSRP
~$15,000 MSRP
NVIDIA GH200 96GBBiggest leap