All writing

Technology evolution

The A20 Pro story is really about heat

Apple’s new 2-nanometre phone chip is impressive, but memory bandwidth, packaging, and sustained performance reveal where mobile computing is going.

The easiest way to market a new chip is with a bigger performance number. The more interesting engineering is often around it. In the iPhone 18 Pro, Apple has paired its A20 Pro chip with more memory bandwidth, a new package, and a redesigned cooling system. That combination says more about the future of phones than one peak benchmark.

According to Apple's announcement, A20 Pro is built on a 2-nanometre process. It has a six-core CPU, seven-core GPU, and two 16-core Neural Engines. Apple claims 50 per cent more memory bandwidth than A19 Pro, up to 40 per cent faster graphics, and twice the AI compute. These are vendor measurements, so independent testing still matters.

Fast for a moment is not the same as fast for a task

A phone has little room to move heat and a battery that must last all day. It can reach a high speed briefly, then reduce power as temperature rises. That is why sustained performance matters for games, video, photography, and on-device models that run longer than a single command.

Apple says its new package places the silicon beside the memory, removes memory from the chip's thermal path, and connects the processor directly to a vapour chamber with three times the previous surface area. The company claims up to a 40 per cent sustained-performance gain over the prior generation.

The exact result will vary by workload and environment. The design direction is clear: mobile progress now depends on moving data and heat as carefully as it depends on adding compute.

AI makes bandwidth visible

On-device AI models repeatedly move weights and intermediate results through memory. A faster neural engine helps only if the rest of the system can keep it supplied. More bandwidth can reduce waiting, while local processing can lower latency and keep some personal data on the device.

That does not mean every AI feature should run locally. Large models may still need cloud capacity, and a hybrid system must explain which path it takes. The useful metric is not a TOPS number by itself. It is whether a feature responds quickly, works reliably, preserves battery, and gives the user a clear privacy boundary.

The chip is becoming a collection of specialised systems

A20 Pro sits beside Apple's N1 networking chip and C2 cellular modem. This reflects a wider shift away from treating the processor as one general-purpose engine. Phones increasingly coordinate specialised compute for graphics, neural workloads, connectivity, image processing, and security.

For buyers, independent battery and sustained-load tests will be more revealing than the launch graph. For designers and developers, the opportunity is subtler: local compute is becoming capable enough to make software feel immediate and private, but only when the whole system is designed around it.

Sources & further reading