None
EN
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
['Jeffrey Kampman', 'Senior Analyst', 'Layer Van-Ds .Vds-Author Font-Size Var --Text-Base', 'Font-Size Var --Vds-Author-Font-Size', 'Var --Text-Base', 'Line-Height Var --Text-Base--Line-Height', 'Line-Height Var --Vds-Author-Line-Height', 'Var --Text-Base--Line-Height', 'Letter-Spacing Var --Tracking-Normal', 'Letter-Spacing Var --Vds-Author-Letter-Spacing']
Latest from Tom's Hardware
Why focus so much on memory bandwidth for local LLM performance to begin with?
The higher-end version uses the M3 Ultra SoC with 819GB/s of memory bandwidth across both the 28-CPU-core, 60-GPU-core mode and the 32-CPU-core, 80-GPU-core model.
The M4 Max version of the Mac Studio encompasses two versions with greater divergence in specs than just core counts.
Assuming that the same organization has carried through to current Apple GPUs, the 40-core M4 Max GPU would have 5,120 execution units or shader ALUs.
(Image credit: Tom's Hardware)The M4 Max Studio we tested has a whopping four Thunderbolt 5 ports on its back panel with support for data rates up to 120Gbps.