RTX 5090 32GBBudget pickB82 tok/s decode
~$1,999 MSRP
Can it run?
Tight fit
Using Q4_K_M in Ollama
Fit status
Tight fit
Decode
18.2 tok/s
TTFT
10635 ms
Safe context
18K
Memory
23.5 GB / 25.9 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | C | Runs with offload (needs ~0.7 GB host RAM) | 17.4 tok/s | 16194 ms | 30K |
| Chat | C | Tight fit | 17.6 tok/s | 5985 ms | 10K |
| Coding | C | Tight fit | 18.2 tok/s | 10635 ms | 18K |
| RAG | C | Runs with offload (needs ~0.7 GB host RAM) | 17.4 tok/s | 20242 ms | 30K |
| Reasoning | C | Tight fit | 17.6 tok/s | 12968 ms |
How Devstral Small 1.1 (24B params) fits at each quantization level on MacBook Pro M4 Max 36GB (25.9 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 9.4 GB | Low | D37 |
Q3_K_S | 3 | 11.8 GB | Low | D39 |
NVFP4 | 4 |
ollama run devstral-small-2507huggingface-cli download devstral-small-2507Upgrade options
RTX 5090 32GBBudget pick~$1,999 MSRP
~$2,499 MSRP
| 18K |
13.4 GB |
| Medium |
| C40 |
Q4_K_M | 4 | 14.6 GB | Medium | C41 |
Q5_K_M | 5 | 17.3 GB | High | C44 |
Q6_KBest for your GPU | 6 | 19.7 GB | High | C44 |
Q8_0 | 8 | 25.7 GB | Very High | C44 |
F16 | 16 | 49.2 GB | Maximum | F0 |
~$2,499 MSRP