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
23.3 tok/s
TTFT
8305 ms
Safe context
17K
Memory
22.0 GB / 24.0 GB
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Agentic Coding | D | Runs with offload (needs ~1 GB host RAM) | 22.0 tok/s | 12816 ms | 30K |
| Chat | C | Tight fit | 21.2 tok/s | 4970 ms | 10K |
| Coding | C | Tight fit | 23.3 tok/s | 8305 ms | 17K |
| RAG | D | Runs with offload (needs ~1 GB host RAM) | 22.0 tok/s | 16020 ms | 30K |
| Reasoning | C | Tight fit | 23.3 tok/s | 9815 ms |
How Devstral Small 2 24B Instruct (24B params) fits at each quantization level on RTX 4500 Ada 24GB (24.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 9.4 GB | Low | D38 |
Q3_K_S | 3 | 11.8 GB | Low | D40 |
NVFP4 | 4 |
ollama run devstral-small-2-24bhuggingface-cli download devstral-small-2-24bUpgrade options
RTX 5090 32GBBudget pick~$1,999 MSRP
~$2,499 MSRP
| 17K |
13.4 GB |
| Medium |
| C41 |
Q4_K_M | 4 | 14.6 GB | Medium | C42 |
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 |
RTX PRO 4500 Blackwell 32GBNVIDIA upgrade