For those not in the know, Apple has attracted a lot of criticism for missing the AI revolution and is now a couple of years behind its rivals. The Tame Apple Press had been hoping that Apple's M5 chip would be the start of it catching up.
If you believe the press release, Jobs' Mob seems to be claiming that it has not only caught up, it is now well ahead of its rivals. If you believe that I have an iron radio tower in Paris I would like to sell you.
Job’s Mob claims sweeping CPU, GPU and neural engine revs, promising “revolutionary” creation and pro workflows. Built on a third-generation 3nm process, the M5 trots out a new 10-core GPU, each with a neural network accelerator, and boasts four times the peak GPU compute of M4.
Apple’s slides talk up third-generation ray tracing and say graphics can improve by up to 45 per cent in heavy creative or gaming loads, which is marketing until independently tested.
On the CPU side, M5 has up to 10 cores with six efficiency cores and up to four performance cores, and claims about 15 per cent better multithreaded performance than M4.
There is the usual “industry-leading energy efficiency” line with nothing comparable, which will please admen more than engineers trying to validate it across workloads and thermals.
Apple, senior vice president of hardware technology, Johny Srouji said: "M5 is a major step forward in the AI performance of Apple chips. With the introduction of neural network accelerators in GPUs, combined with faster neural engines and higher memory bandwidth, M5 delivers unprecedented performance for MacBook Pro, iPad Pro, and Apple Vision Pro."
That lands alongside talk of a GPU architecture “optimised for AI work” where every core handles inference and rendering, which reads well on a slide but needs real benches.
Apple claims third-generation ray tracing and second-generation dynamic caching for 3D modelling and games, while name-checking Vision Pro’s micro-OLED at 120Hz for smoother visuals.
The spiel leans on tight ties with Metal Performance Shaders, Core ML and Metal 4, letting developers tug on tensor bits, which is convenient if you live entirely in Apple’s walled garden.
M5’s neural engine is billed as faster and thriftier, with system features like Apple Intelligence shoved to the front as beneficiary, though on-device latency claims arrive without hard numbers.
Vision Pro party tricks such as 2D scene generation in Photos and auto “self-portrait” puffery are said to run quicker, and Apple Foundation Models are pitched as snappier for devs.
Unified memory bandwidth is quoted at 153GB/s, nearly 30 per cent up on M4 and more than two times M1, which might help big models if the rest of the system keeps up.
The unified memory pool lets CPU, GPU and neural bits poke the same data to avoid shuffling, while configurations top out at 32GB RAM, which will cramp some pro AI and video workflows.
Apple finishes by crowning M5 the cornerstone of its AI push, which is now so delayed and behind rivals it needs it.
