MacBook Pro M4 32GBBest value
C6.4 tok/s decode
~$799 MSRP
Can it run?
Runs with offload (needs ~0.8 GB host RAM)
Using Q4_K_M in vLLM
Fit status
Runs with offload (needs ~0.8 GB host RAM)
Decode
35.3 tok/s
TTFT
5480 ms
Safe context
15K
Memory
21.3 GB / 20.0 GB
Offload
10%
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | F | Too heavy | 37.2 tok/s | 7570 ms | 26K |
| Chat | C | Runs with offload | 37.2 tok/s | 2839 ms | 8K |
| Coding | C | Runs with offload (needs ~0.8 GB host RAM) | 35.3 tok/s | 5480 ms | 15K |
| RAG | F | Too heavy | 37.2 tok/s | 9463 ms | 26K |
| Reasoning | C | Runs with offload (needs ~0.8 GB host RAM) | 35.3 tok/s | 6476 ms | 15K |
How Codestral 2 25.08 (22B params) fits at each quantization level on RTX A4500 20GB (20.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.6 GB | Low | D39 |
Q3_K_S | 3 | 10.8 GB | Low | C41 |
NVFP4 | 4 | 12.3 GB | Medium | C43 |
Q4_K_MBest for your GPU | 4 | 13.4 GB | Medium | C44 |
Q5_K_M | 5 | 15.8 GB | High | C44 |
Q6_K | 6 | 18.0 GB | High | C44 |
Q8_0 | 8 | 23.5 GB | Very High | F0 |
F16 | 16 | 45.1 GB | Maximum | F0 |
huggingface-cli download codestral-2-25.08Upgrade options
~$799 MSRP
~$999 MSRP
RTX 5090 32GBBiggest leap~$1,999 MSRP
RTX PRO 4500 Blackwell 32GBNVIDIA upgrade