RX 7900 XTX 24GBBudget pick
C47.2 tok/s decode
~$999 MSRP
Can it run?
Runs with offload (needs ~0 GB host RAM)
Using Q4_K_M in Ollama
Fit status
Runs with offload (needs ~0 GB host RAM)
Decode
9.6 tok/s
TTFT
20247 ms
Safe context
16K
Memory
23.0 GB / 23.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | D | Very compromised (needs ~2.1 GB host RAM) | 9.2 tok/s | 30740 ms | 28K |
| Chat | C | Tight fit | 9.6 tok/s | 11043 ms | 9K |
| Coding | C | Runs with offload (needs ~0 GB host RAM) | 9.6 tok/s | 20247 ms | 16K |
| RAG | D | Very compromised (needs ~2.1 GB host RAM) | 9.2 tok/s | 38425 ms | 28K |
| Reasoning | C | Runs with offload (needs ~0 GB host RAM) | 9.6 tok/s | 23928 ms | 16K |
How Devstral Small 2 24B Instruct (24B params) fits at each quantization level on MacBook Pro M2 Pro 32GB (23.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 9.4 GB | Low | D38 |
Q3_K_S | 3 | 11.8 GB | Low | C40 |
NVFP4 | 4 | 13.4 GB | Medium | C42 |
Q4_K_M | 4 | 14.6 GB | Medium | C43 |
Q5_K_MBest for your GPU | 5 | 17.3 GB | High | C44 |
Q6_K | 6 | 19.7 GB | High | C44 |
Q8_0 | 8 | 25.7 GB | Very High | F0 |
F16 | 16 | 49.2 GB | Maximum | F0 |
ollama run devstral-small-2-24bhuggingface-cli download devstral-small-2-24bUpgrade options
~$999 MSRP
RTX 3090 24GBBest value~$1,499 MSRP
RTX 5090 32GBBiggest leap~$1,999 MSRP
~$2,499 MSRP