Apple announced the M6 chip last week, just ahead of its September iPhone event, providing the clearest indication yet of what the A20 will bring to the iPhone 18 lineup.
Generations of Apple silicon share much of their architecture between the A and M series, so a new A series chip normally functions as a partial preview of the M series chip that follows it. This year, the order was reversed; The M6 was revealed on August 25, ahead of the A20 Pro expected in the iPhone 18 Pro this month.
Recent reports place the A20 Pro up to 18% faster and up to 30% more power efficient than the A19 Pro, which is broadly similar to the M6 compared to the M5.
Apple’s first 2nm iPhone chip
The M6 is Apple’s first chip built on TSMC’s 2nm (N2) process, with the second almost certainly being the A20. Being a smaller manufacturing process, the N2 has a higher transistor density, allowing Apple to fit more performance into a smaller area.
Apple has historically launched new nodes on the A-series, since iPhone volume validates returns, so introducing the N2 in an $899 Mac mini represents a change. As such, reports that the A20 and A20 Pro will upgrade to N2 are now supported by a shipping product rather than analyst predictions. The consequence for the iPhone event is that the claim “Apple’s first 2nm chip” is no longer available.
Double neural engine?
The iPhone’s neural engine has remained at 16 cores and around 35 trillion operations per second since the A17 Pro, across three generations. The M6 introduces a 16-core dual neural engine, which Apple claims system frameworks can simultaneously handle up to 2x the peak computing of previous generations.
The change corresponds to the reported move to WMCM packaging, which integrates memory on the same slice as the CPU, GPU and Neural Engine to shorten the path between them. It is possible that die surface area and thermal constraints limit the use of a dual motor in the A20 Pro.
GPU AI gains are now incremental
Apple said the M5 offers more than 4x the maximum GPU compute for AI compared to the M4, reflecting the introduction of neural accelerators in each GPU core. The M6 is rated almost 30% higher than the M5.
The A19 Pro saw a comparable 4x improvement over the iPhone 17 Pro last September for the same architectural reason. The second-generation neural accelerators therefore appear to represent a refinement rather than another radical change, suggesting that the A20 Pro won’t repeat the A19 Pro’s flagship figure.
Graphics improvements should trickle down
The M6 adds 50% increased geometry rates, updated core shader architecture, revised dynamic caching, and hardware-accelerated ray tracing. These are architectural changes and comparable graphic works have historically appeared in the A and M series within the same generation.
Memory bandwidth suggests modest A20 gain
Memory bandwidth determines how quickly data flows between memory and the processor, and is the main constraint when running large language models on the device. The M6 offers up to 170 GB/s, which Apple describes as a 10% increase over the M5’s 153.6 GB/s.
This increase is small enough to indicate that Apple is using a faster quality of the same LPDDR5X memory rather than moving to the new LPDDR6 standard, which would have allowed a much bigger jump. Reports indicate that Apple is sourcing high-speed LPDDR5X from Samsung for the iPhone 18 lineup, and the M6 is consistent with that.
The implication for the A20 is that memory bandwidth is unlikely to improve dramatically. The A19 Pro is rated at 76.8 GB/s, and a similar 10% increase would bring the A20 Pro to around 85 GB/s. Any significant gain in how quickly the iPhone 18 Pro loads and runs models on the device is therefore more likely to come from WMCM packaging shortening the distance data travels than from the memory itself.
AV1 encoding remains absent
The M6 Media Engine supports H.264 hardware acceleration, HEVC, ProRes and ProRes RAW, with AV1 decoding but no AV1 encoding, matching the M5. The omission of a new process node in a desktop chip makes its inclusion in the A20 Pro very unlikely.
