RX 7900 XTX 24GBBudget pick
C54 tok/s decode
~$999 MSRP
Can it run?
Tight fit
Using Q4_K_M in Ollama
Fit status
Tight fit
Decode
6.1 tok/s
TTFT
31795 ms
Safe context
18K
Memory
20.7 GB / 23.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs with offload (needs ~0.5 GB host RAM) | 6.7 tok/s | 42273 ms | 31K |
| Chat | C | Tight fit | 6.7 tok/s | 15657 ms | 10K |
| Coding | C | Tight fit | 6.1 tok/s | 31795 ms | 18K |
| RAG | C | Runs with offload (needs ~0.5 GB host RAM) | 6.7 tok/s | 52841 ms | 31K |
| Reasoning | C | Tight fit | 6.7 tok/s | 33923 ms |
How Codestral 21B Pruned i1 (21B params) fits at each quantization level on MacBook Pro M4 32GB (23.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 8.2 GB | Low | D36 |
Q3_K_S | 3 | 10.3 GB | Low | D38 |
NVFP4 | 4 |
Upgrade options
~$999 MSRP
RTX 3090 24GBBest value~$1,499 MSRP
RTX 5090 32GBBiggest leap~$1,999 MSRP
| 18K |
11.8 GB |
| Medium |
| D40 |
Q4_K_M | 4 | 12.8 GB | Medium | C41 |
Q5_K_M | 5 | 15.1 GB | High | C43 |
Q6_KBest for your GPU | 6 | 17.2 GB | High | C44 |
Q8_0 | 8 | 22.5 GB | Very High | C44 |
F16 | 16 | 43.1 GB | Maximum | F0 |
~$2,499 MSRP