AMD Instinct MI350X 288GBBudget pick
C133 tok/s decode
~$8,000 MSRP
Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
5.9 tok/s
TTFT
32704 ms
Safe context
4K
Memory
71.4 GB / 92.2 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Tight fit | 5.5 tok/s | 51533 ms | 4K |
| Chat | C | Runs well | 5.5 tok/s | 19325 ms | 4K |
| Coding | C | Runs well | 5.9 tok/s | 32704 ms | 4K |
| RAG | C | Tight fit | 5.5 tok/s | 64416 ms | 4K |
| Reasoning | C | Runs well | 5.5 tok/s | 41870 ms | 4K |
How Qwen 2.5 Math 72B (72B params) fits at each quantization level on MacBook Pro M3 Max 128GB (92.2 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 28.1 GB | Low | D36 |
Q3_K_S | 3 | 35.3 GB | Low | D38 |
NVFP4 | 4 |
ollama run qwen-2.5-math-72bhuggingface-cli download qwen-2.5-math-72bUpgrade options
~$8,000 MSRP
~$15,000 MSRP
NVIDIA GH200 96GBBiggest leap40.3 GB |
| Medium |
| D39 |
Q4_K_M | 4 | 43.9 GB | Medium | D40 |
Q5_K_M | 5 | 51.8 GB | High | C41 |
Q6_KBest for your GPU | 6 | 59.0 GB | High | C43 |
Q8_0 | 8 | 77.0 GB | Very High | C44 |
F16 | 16 | 147.6 GB | Maximum | F0 |