AMD Instinct MI250X 128GBBudget pick
C155 tok/s decode
~$15,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
21.4 tok/s
TTFT
9064 ms
Safe context
23K
Memory
64.6 GB / 92.2 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 21.4 tok/s | 13184 ms | 46K |
| Chat | C | Runs well | 21.4 tok/s | 4944 ms | 11K |
| Coding | C | Runs well | 21.4 tok/s | 9064 ms | 23K |
| RAG | C | Runs well | 21.4 tok/s | 16479 ms | 46K |
| Reasoning | C | Runs well | 21.4 tok/s | 10712 ms | 23K |
How Qwen3-Coder-Next (80B params) fits at each quantization level on MacBook Pro M4 Max 128GB (92.2 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 31.2 GB | Low | D37 |
Q3_K_S | 3 | 39.2 GB | Low | D39 |
NVFP4 | 4 | 44.8 GB | Medium | D40 |
Q4_K_M | 4 | 48.8 GB | Medium | C41 |
Q5_K_M | 5 | 57.6 GB | High | C43 |
Q6_KBest for your GPU | 6 | 65.6 GB | High | C45 |
Q8_0 | 8 | 85.6 GB | Very High | C45 |
F16 | 16 | 164.0 GB | Maximum | F0 |
ollama run qwen-3-coder-nexthuggingface-cli download qwen-3-coder-nextUpgrade options
~$15,000 MSRP
NVIDIA GH200 96GBBiggest leap