StudioBenchmark: Same 32 Models on M2 Pro, M3 Ultra, M4 Pro and M5 Max
Original title: Benchmark Same 32 models on M2 Pro, M3 Ultra M4 Pro and M5 Max
A Redditor ran the same MLX benchmark on four Macs with identical settings, measuring decode speed for several models and Qwen-Image 2.1 generation. MoE models run as fast or faster on the M5 Max than the M3 Ultra since few params are active, while the dense 27B favors the Ultra's bandwidth; image generation is ~7x faster on M5 Max vs M4 Pro.
Why it matters
The author benchmarked decode speeds under identical conditions across four Macs, showing how MoE models depend more on GPU generation while dense models remain bandwidth-bound.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.