AMD Instinct MI350X 288GBBudget pick
C435.2 tok/s decode
~$8,000 MSRP
Can it run?
Runs well
Using Q4_K_M in vLLM
Fit status
Runs well
Decode
41.5 tok/s
TTFT
4665 ms
Safe context
63K
Memory
46.9 GB / 184.3 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 41.5 tok/s | 6786 ms | 117K |
| Chat | C | Runs well | 41.5 tok/s | 2545 ms | 33K |
| Coding | C | Runs well | 41.5 tok/s | 4665 ms | 63K |
| RAG | C | Runs well | 41.5 tok/s | 8482 ms | 117K |
| Reasoning | C | Runs well | 41.5 tok/s | 5513 ms | 63K |
How Codestral 2 25.08 (22B params) fits at each quantization level on Mac Studio M3 Ultra 256GB (184.3 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.6 GB | Low | D30 |
Q3_K_S | 3 | 10.8 GB | Low | D31 |
NVFP4 | 4 | 12.3 GB | Medium | D31 |
Q4_K_M | 4 | 13.4 GB | Medium | D31 |
Q5_K_M | 5 | 15.8 GB | High | D31 |
Q6_K | 6 | 18.0 GB | High | D31 |
Q8_0 | 8 | 23.5 GB | Very High | D32 |
F16Best for your GPU | 16 | 45.1 GB | Maximum | D35 |
huggingface-cli download codestral-2-25.08Upgrade options
~$8,000 MSRP
~$15,000 MSRP
H100 NVL 188GBBiggest leap