.png)
Most new chips arrive with a faster clock and a bigger number on the box. The M5 did something more interesting. Apple tucked a small block of matrix-math hardware, called a Neural Accelerator, inside every GPU core, and says that change gives it over 4x the peak GPU compute for AI compared with the M4.
That was October 2025. A year on, the idea has spread to the MacBook Air, the M5 Pro and M5 Max laptops and an M5 Ultra Mac Studio, and the brand-new M6 builds on it. Here is what the accelerators actually do, what Apple has confirmed, and what U.S. buyers should think about before spending money.
Checked October 30, 2026 Every performance number below is Apple’s own claim unless stated otherwise.
The Quick Answer
A Neural Accelerator is a matrix-math unit built into each GPU core. It speeds up the big multiplication jobs behind language models and image generators, so a Mac can do more AI work on the chip instead of leaning on the cloud.
The 4x headline is Apple’s, and it is a peak figure. Apple’s own language-model test shows a nuanced picture: about 3.3x to 4.1x faster before a model starts answering, but only 19 to 27 percent faster once it is generating text.
The M5 is not obsolete yet. The newer M6 arrives in a 14-inch MacBook Pro on November 6, but Apple’s touchscreen MacBook Pro, announced October 28, still runs on the M5 Pro and M5 Max.
Neural Accelerators vs. the Neural Engine
People mix these two up, and it is easy to see why. The Neural Engine is a separate 16-core block that Apple pitches as the efficient home for things like Apple Intelligence and Image Playground. The Neural Accelerators sit inside the GPU, one per core, and handle dedicated matrix multiplication, the heavy lifting behind large language models, according to Apple’s machine learning team.
The placement matters for software. Apps that already use Core ML, Metal Performance Shaders or Metal 4 can pick up gains automatically, Apple says, and developers can program the accelerators directly through Tensor APIs in Metal 4. Apple’s open-source MLX framework taps them too, though the speedups need macOS 26.2 or later.
What the M5 Launched With
Apple unveiled the M5 on October 15, 2025, in the 14-inch MacBook Pro, iPad Pro and Apple Vision Pro. It is a third-generation 3-nanometer chip with a 10-core GPU and a CPU of up to 10 cores, four performance and six efficiency. Apple claims up to 15 percent faster multithreaded performance and up to 45 percent better ray-tracing graphics than the M4.
Memory got a lift as well. Bandwidth rose nearly 30 percent to 153GB/s, and the chip supports up to 32GB. All of this is from Apple’s own preproduction testing.
What Apple’s Own Benchmark Shows
Apple’s research team compared a 24GB MacBook Pro with M5 against a similarly configured M4, feeding six language models a 4,096-token prompt. Time to first token, the pause before the model starts answering, improved by roughly 3.3x to 4.1x. A 1024-by-1024 image from a 12-billion-parameter FLUX model came out more than 3.8x faster.
Once the first word appears, the story is gentler. Later tokens ran 19 to 27 percent faster, which Apple credits to memory bandwidth climbing from 120GB/s to 153GB/s. The pattern is worth remembering: reading a long prompt is limited by compute, so accelerators help a lot, while producing each following word is limited by memory speed. In daily use, long documents and code should feel snappier, and chat speed improves more modestly. Other apps will vary.
How the Idea Scaled Up
The design did not stay in the base chip. On March 3, 2026, Apple introduced the M5 Pro and M5 Max, which join two 3-nanometer dies into one chip using what Apple calls Fusion Architecture. Both have an 18-core CPU, and because every GPU core carries an accelerator, the count grows with the GPU. Apple says both deliver over 4x the previous generation’s peak GPU compute for AI.
Bandwidth: 153GB/s
Max memory: 32GB
Bandwidth: up to 307GB/s
Max memory: 64GB
Bandwidth: up to 614GB/s
Max memory: 128GB
Bandwidth: up to 170GB/s
Max memory: 32GB
Specs as Apple lists them in the releases linked in this section. Each GPU core carries one Neural Accelerator.
The base M5 also reached the MacBook Air, which launched at $1,099 for the 13-inch model with 512GB of storage. MacRumors reports that Apple raised that price by $200 in June, pointing to higher component costs.
For the Max tier, Apple’s pitch is capacity. Its October 28 release says M5 Max owners can run language models with hundreds of billions of parameters entirely on the device. That says what the hardware can hold, not how fast it runs on a typical laptop.
At the top, the Mac Studio announced August 25 offers the M5 Max from $2,499 and a new M5 Ultra from $5,499. Apple says the Ultra delivers up to 4.3x the peak AI compute of the M3 Ultra. Coverage of the launch said a 512GB memory option was due in late October, and it is not yet confirmed that deliveries have begun.
Where the M6 Fits
The newest step landed this week. On October 27, Apple announced a 14-inch MacBook Pro with M6, starting at $1,999 and arriving November 6. Its 12-core GPU keeps an accelerator in every core, and Apple pairs it with what it calls a Dual 16-core Neural Engine.
Against the M5, Apple claims up to 25 percent faster AI workloads such as prompt processing, up to 1.9x faster Neural Engine performance and up to 1.3x faster time to first token. Those are real gains, but smaller than the 4x leap Apple claimed when the accelerators first appeared. My read, and it is only that, is that the M6 looks like refinement of the M5 idea rather than a second revolution.
Here is the twist. Apple’s touchscreen MacBook Pro uses the M5 Pro and M5 Max, not the M6. Prices start at $2,899 for the 14-inch M5 Pro and climb to $4,799 for the 16-inch M5 Max, with availability beginning November 6. For now, M5-generation silicon is still Apple’s top laptop option.
What U.S. Buyers Should Weigh
Bigger models need more RAM, and bandwidth sets the pace of word-by-word generation. In Apple’s test, a 24GB machine held an 8-billion-parameter model at full precision, or a 30-billion-parameter mixture-of-experts model compressed to 4-bit, under 18GB.
Apple’s 2026 pricing has shifted, as the MacBook Air shows. The base 14-inch M5 MacBook Pro debuted at $1,599, but it is not confirmed whether Apple still sells it beside the $1,999 M6 model.
Accelerator gains need apps and frameworks that use them. If you mostly use built-in Apple Intelligence features, you may notice less than someone running local models in LM Studio or MLX.
Apple’s numbers are “up to” results from its own labs, compared only with its own earlier chips.
What We Still Don’t Know
- How the M5 and M6 compare across real apps in independent reviews.
- Whether the base M5 MacBook Pro is still sold alongside the M6 model.
- Whether the 512GB Mac Studio has started shipping.
- How Apple’s accelerators perform against discrete GPUs from other vendors.
Quick FAQ
Sources
- Apple Newsroom: Apple unleashes M5
- Apple Newsroom: M5 Pro and M5 Max
- Apple Machine Learning Research: Exploring LLMs with MLX and the Neural Accelerators in the M5 GPU
- Apple Newsroom: 14-inch MacBook Pro with M6
- Apple Newsroom: MacBook Pro with PROLED XDR Display
- Apple Newsroom: Mac Studio with M5 Max and M5 Ultra
- MacTech: Mac Studio pricing from Apple’s release
- Production Expert: M6 and M5 Ultra availability
- MacRumors: MacBook Air guide (June price increase)
- Apple World Today: MacBook Air with M5 pricing
- Tom’s Guide: M5 MacBook Pro launch price
- Images: free to use under the Unsplash License. Photos by Brecht Corbeel, Bermix Studio and Thorium.
.png){kind=logo}
No comments:
Post a Comment