AMD Instinct MI350X 288GBBudget pick
C206.8 tok/s decode
~$8,000 MSRP
Can it run?
Tight fit
Using Q4_K_M in Ollama
Fit status
Tight fit
Decode
8.5 tok/s
TTFT
22783 ms
Safe context
18K
Memory
84.2 GB / 92.2 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Tight fit | 8.5 tok/s | 33138 ms | 34K |
| Chat | C | Tight fit | 8.5 tok/s | 12427 ms | 9K |
| Coding | C | Tight fit | 8.5 tok/s | 22783 ms | 18K |
| RAG | C | Tight fit | 8.5 tok/s | 41423 ms | 34K |
| Reasoning | C | Tight fit | 8.5 tok/s | 26925 ms | 18K |
How Llama 4 Scout 17B 16E (109B params) fits at each quantization level on MacBook Pro M3 Max 128GB (92.2 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 42.5 GB | Low | D39 |
Q3_K_S | 3 | 53.4 GB | Low | C42 |
NVFP4 | 4 | 61.0 GB | Medium | C44 |
Q4_K_MBest for your GPU | 4 | 66.5 GB | Medium | C44 |
Q5_K_M | 5 | 78.5 GB | High | C44 |
Q6_K | 6 | 89.4 GB | High | C44 |
Q8_0 | 8 | 116.6 GB | Very High | F0 |
F16 | 16 | 223.5 GB | Maximum | F0 |
ollama run llama-4-scout-17b-16ehuggingface-cli download llama-4-scout-17b-16eUpgrade options
~$8,000 MSRP
~$15,000 MSRP
Gaudi 3 128GBBiggest leap