M3 Ultra 256GB vs RTX 5090
For running local LLMs · 67 reports across 12 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 | M3 Ultra 256GB | RTX 5090 |
|---|---|---|
| Qwen3.8up to 2400B | — | 220.0n=35 |
| DeepSeek V4 Flashup to 285B | — | 120.0n=10 |
| Qwen3.6up to 35B | — | 215.1n=6 |
| Gemma 4up to 31B | — | 578.0n=4 |
| Museup to 30B | — | 253.0n=4 |
| GLM-5.3up to 321B | 37.4n=1 | 24.5n=1 |
| GLM-5.2up to 744B | — | —n=1 |
| Kimi K2.5 | — | —n=1 |
| Minimax M3 | 17.3n=1 | — |
| Qwen2.5up to 32B | — | —n=1 |
| Qwen3up to 235B | — | 159.0n=1 |
| Qwen3.5up to 122B | — | —n=1 |