EP 192: Talking Apple Silicon with Apple's Tom Boger…

PodcastThe CircuitTom BogerSep 15

Apple describes the A20 Pro's performance and thermal design, including higher CPU and GPU speeds, a larger Neural Engine, and packaging intended to improve heat extraction and bandwidth. Its AI approach combines on-device processing with Private Cloud Compute, while assigning workloads across the Neural Engine, GPU, and CPU. Speakers also point to growing demand for local AI, specialized models, and silicon designed alongside products to accommodate future capabilities.

Semiconductors Artificial Intelligence AI Infrastructure

Preview · 11 more points in the app

Read the Full Summary

Key Takeaways

Apple says the A20 Pro is 20% faster on CPU and 40% faster on GPU, with higher transistor density enabling more Neural Engine capacity, GPU cores, memory channels, and bandwidth. Its DRAM placement beside the SoC and vapor-chamber thermal design are intended to improve heat extraction and sustain performance for AI and graphics workloads, while Apple also credits design work with reducing power.

Apple says it designs its silicon and products together, allowing it to plan bandwidth for future camera and AI capabilities and leave headroom for software updates. That approach is reflected in the A20 Pro's added memory bandwidth, which Apple says supports AI and graphics workloads.

Apple describes AI work as distributed across the Neural Engine, GPU, and CPU according to performance and efficiency needs, with its Core AI framework selecting whether to use one or both Neural Engine core sets. The dual 16-core engines can run separate workloads or split one, and some tasks can operate at lower frequency and voltage when maximum performance is unnecessary.

TECHNOLOGY

A20 Pro Gains — Apple says the A20 Pro's higher transistor density enabled a doubled Neural Engine, an extra GPU core, more memory channels and bandwidth, a CPU 20% faster than its predecessor, and a GPU 40% faster. The speaker also credits design work, not just process technology, with reducing power.

“We doubled the neural engine. We added extra GPU core. We added extra memory channels for memory, more memory bandwidth into the chip…”—

Vapor Chamber Design — Apple says custom thermal solutions in the iPhone 18 Pro, Pro Max, and Duo place the chip on a vapor chamber. The design aims to improve sustained performance during AI workloads and gaming, rather than focusing only on peak performance.

“Using that custom package, we were able then to put the chip right on the vapor chamber to allow us to get the best sustained performance throughout…”—

Efficient Mac AI — Apple says the M6 Mac mini's performance per watt makes it well suited to run continuously on a desk for agentic AI workloads. The speaker also says four Mac Studios can run frontier models from a common wall outlet.

“And one of the reasons it's so important these days is just the rise of Agenetic AI and the use of a Mac mini where it's sitting there on your…”—

Apple Silicon Roadmap

Ben introduces Apple’s product and silicon leaders, then frames the discussion around how process advances and vertical integration translate into user-facing benefits.

A20 Pro Performance

Sri attributes the A20 Pro’s faster CPU and GPU, doubled Neural Engine, and added memory bandwidth to both two-nanometer density and design work focused on efficiency.

Thermals And Sustained Speed

The guests explain how power efficiency and thermal design preserve sustained performance for Mac AI workloads, iPhone gaming, and other demanding tasks.

A20 Pro Packaging

The speakers describe moving DRAM beside the A20 Pro to improve heat extraction and bandwidth, while coordinating board, vapor-chamber, and material changes to boost sustained performance.

Dual Neural Engine

Sri says the paired Neural Engines can split or run workloads in parallel for performance and efficiency, while Apple’s frameworks select the appropriate configuration for apps and features.

Developers And On-Device AI

The guests expect improved compute and sustained performance to give developers room for richer experiences, and describe routing complex AI workloads between devices and Private Cloud Compute.

Local AI And Specialized Models

The speakers cite customer demand for local AI driven by token costs and privacy, alongside an industry shift toward selecting specialized models for different tasks.