M5 Max Benchmarks Crush Competitors

The M5 Pro and M5 Max chips are the first in the series to adopt Apple's "Fusion Architecture," connecting two separate dies into a single SoC. This multi-die approach, manufactured on TSMC's third-generation 3nm (N3P) process, moves away from the previous monolithic design. Architecturally, the high-end M5 chips abandon the traditional efficiency-core design. The 18-core M5 Max instead utilizes six high-performance "super cores" alongside 12 all-new "performance cores" optimized specifically for power-efficient multithreaded workloads. The most significant change for AI workloads is the introduction of a Neural Accelerator embedded directly into each of the M5 Max's up-to-40 GPU cores. This works in tandem with the updated 16-core Neural Engine to deliver what Apple claims is over four times the peak GPU compute for AI compared to the M4 generation. For developers, these new Neural Accelerators can be programmed directly using new Tensor APIs in Metal 4. Applications built with existing frameworks like Core ML and Metal Performance Shaders will gain performance automatically without code changes. This performance is supported by a substantial memory subsystem, with the M5 Max offering up to 614GB/s of memory bandwidth and configurable with up to 128GB of unified memory. This makes running large language models, such as 70-billion-parameter models, locally on a laptop a practical reality for the first time. The Geekbench 6 multi-core scores not only surpass mobile competitors but also high-end desktop CPUs like the 96-core AMD Ryzen Threadripper Pro 9995WX in bursty workloads. However, Geekbench is a short-duration test and may not fully represent performance in sustained, professional applications. Beyond core performance, the new silicon includes hardware-accelerated AV1 decoding in its Media Engine and support for the Thunderbolt 5 standard for high-speed peripherals.