AMD Instinct MI350X 288GBBudget pick
C234.1 tok/s decode
~$8,000 MSRP
Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
22.3 tok/s
TTFT
8673 ms
Safe context
28K
Memory
103.7 GB / 184.3 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 22.3 tok/s | 12615 ms | 56K |
| Chat | C | Runs well | 22.3 tok/s | 4731 ms | 14K |
| Coding | C | Runs well | 22.3 tok/s | 8673 ms | 28K |
| RAG | C | Runs well | 22.3 tok/s | 15769 ms | 56K |
| Reasoning | C | Runs well | 22.3 tok/s | 10250 ms | 28K |
How Mistral Small 4 119B (119B params) fits at each quantization level on Mac Studio M3 Ultra 256GB (184.3 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 46.4 GB | Low | D35 |
Q3_K_S | 3 | 58.3 GB | Low | D36 |
NVFP4 | 4 | 66.6 GB | Medium | D37 |
Q4_K_M | 4 | 72.6 GB | Medium | D38 |
Q5_K_M | 5 | 85.7 GB | High | D40 |
Q6_K | 6 | 97.6 GB | High | C41 |
Q8_0Best for your GPU | 8 | 127.3 GB | Very High | C44 |
F16 | 16 | 244.0 GB | Maximum | F0 |
huggingface-cli download mistral-small-4-119bUpgrade options
~$8,000 MSRP
~$15,000 MSRP
H100 NVL 188GBBiggest leap