MacBook Pro M4 Pro 24GBBudget pick
C86.2 tok/s decode
~$1,999 MSRP
Can it run?
Runs well
Using Q4_K_M in Ollama
Fit status
Runs well
Decode
162.6 tok/s
TTFT
1190 ms
Safe context
42K
Memory
6.0 GB / 16.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs well | 162.6 tok/s | 1732 ms | 79K |
| Chat | C | Runs well | 162.6 tok/s | 649 ms | 21K |
| Coding | C | Runs well | 162.6 tok/s | 1190 ms | 42K |
| RAG | C | Runs well | 162.6 tok/s | 2165 ms | 79K |
| Reasoning | C | Runs well | 162.6 tok/s | 1407 ms | 42K |
How gemma 3 4b it (4B params) fits at each quantization level on RX 9070 16GB (16.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 1.6 GB | Low | D32 |
Q3_K_S | 3 | 2.0 GB | Low | D32 |
NVFP4 | 4 | 2.2 GB | Medium | D32 |
Q4_K_M | 4 | 2.4 GB | Medium | D33 |
Q5_K_M | 5 | 2.9 GB | High | D33 |
Q6_K | 6 | 3.3 GB | High | D34 |
Q8_0 | 8 | 4.3 GB | Very High | D35 |
F16Best for your GPU | 16 | 8.2 GB | Maximum | C40 |
Upgrade options
~$1,999 MSRP