RTX 3090 vs RTX 5090
For running local LLMs · 150 reports across 15 models
Highest reported generation speed by model family
These maxima come from different community setups. Model size, quantization, GPU count, offloading and concurrency may differ. They do not establish which GPU is faster for the same workload. Check the comparison conditions →
| Model | RTX 3090 | RTX 5090 |
|---|---|---|
| Qwen3.8up to 2400B | 382.0n=51 | 220.0n=35 |
| DeepSeek V4 Flashup to 285B | 60.0n=9 | 120.0n=10 |
| Qwen3.6up to 35B | 170.7n=13 | 215.1n=6 |
| Museup to 30B | 94.0n=1 | 253.0n=4 |
| Gemma 4up to 31B | — | 578.0n=4 |
| GLM-5.2up to 744B | 7.3n=2 | —n=1 |
| GLM-5.3up to 321B | 55.0n=2 | 24.5n=1 |
| Qwen2.5up to 32B | 70.0n=2 | —n=1 |
| Qwen3up to 235B | 7.5n=1 | 159.0n=1 |
| DeepSeek R1 | 2.0n=1 | — |
| DeepSeek V3 | —n=1 | — |
| DeepSeek V4up to 284B | 15.0n=1 | — |
| GLM-4.5-Airup to 106B | —n=1 | — |
| Kimi K2.5 | — | —n=1 |
| Qwen3.5up to 122B | — | —n=1 |