AMD Instinct MI350X 288GBBudget pick
C133 tok/s decode
~$8,000 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
12.7 tok/s
TTFT
15268 ms
Safe context
35K
Memory
84.0 GB / 184.3 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 12.7 tok/s | 22208 ms | 62K |
| Chat | C | Runs well | 12.7 tok/s | 8328 ms | 19K |
| Coding | C | Runs well | 12.7 tok/s | 15268 ms | 35K |
| RAG | C | Runs well | 12.7 tok/s | 27760 ms | 62K |
| Reasoning | C | Runs well | 12.7 tok/s | 18044 ms | 35K |
How Qwen 2.5 72B (72B params) fits at each quantization level on Mac Studio M3 Ultra 256GB (184.3 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 28.1 GB | Low | D33 |
Q3_K_S | 3 | 35.3 GB | Low | D34 |
NVFP4 | 4 | 40.3 GB | Medium | D34 |
Q4_K_M | 4 | 43.9 GB | Medium | D35 |
Q5_K_M | 5 | 51.8 GB | High | D36 |
Q6_K | 6 | 59.0 GB | High | D36 |
Q8_0 | 8 | 77.0 GB | Very High | D38 |
F16Best for your GPU | 16 | 147.6 GB | Maximum | C45 |
ollama run qwen-2.5-72bhuggingface-cli download qwen-2.5-72bUpgrade options
~$8,000 MSRP
~$15,000 MSRP
H100 NVL 188GBBiggest leap