Exploring Apple Silicon’s local AI performance with the Mac Studio and M4 Max — M4 Max beats GB10 and Strix Halo in decode throughput, but memory bandwidth isn't everything

Tom’s Hardware found the Apple M4 Max Mac Studio outperformed Nvidia GB10 and AMD Strix Halo systems in local AI inference, especially decode throughput.

Because LLMs generate tokens sequentially, memory bandwidth is a major bottleneck, and Apple’s large unified memory plus high bandwidth gave it an edge.

The results suggest that for local LLM workloads, memory capacity and bandwidth can matter more than raw compute.

The tested Mac Studio configuration is now harder to buy, with reduced top-end RAM options and longer lead times.