AMD Instinct MI250X 128GBBudget pick
C151.5 tok/s decode
~$15,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
28.2 tok/s
TTFT
6872 ms
Safe context
41K
Memory
35.7 GB / 92.2 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 28.2 tok/s | 9996 ms | 74K |
| Chat | C | Runs well | 28.2 tok/s | 3748 ms | 22K |
| Coding | C | Runs well | 28.2 tok/s | 6872 ms | 41K |
| RAG | C | Runs well | 28.2 tok/s | 12494 ms | 74K |
| Reasoning | C | Runs well | 28.2 tok/s | 8121 ms | 41K |
How Qwen3.5 27B (27B params) fits at each quantization level on Mac Studio M2 Ultra 128GB (92.2 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 10.5 GB | Low | D32 |
Q3_K_S | 3 | 13.2 GB | Low | D33 |
NVFP4 | 4 | 15.1 GB | Medium | D33 |
Q4_K_M | 4 | 16.5 GB | Medium | D33 |
Q5_K_M | 5 | 19.4 GB | High | D34 |
Q6_K | 6 | 22.1 GB | High | D35 |
Q8_0 | 8 | 28.9 GB | Very High | D36 |
F16Best for your GPU | 16 | 55.4 GB | Maximum | C42 |
huggingface-cli download hf-unsloth--qwen3-5-27b-ggufUpgrade options
~$15,000 MSRP
~$15,000 MSRP
NVIDIA GH200 96GBBiggest leap