Robots Atlas>ROBOTS ATLAS
AppleDeployed

Apple M2 Ultra

The Apple M2 Ultra is an Apple Silicon system-on-chip (SoC) introduced on 5 June 2023. It is built by joining two M2 Max dies with UltraFusion โ€” a silicon interposer carrying more than 10,000 signals and providing over 2.5 TB/s of inter-die bandwidth. The chip contains 134 billion transistors and is manufactured on a second-generation 5 nm process.

It features up to a 24-core CPU (16 performance and 8 efficiency cores), a GPU in 60- or 76-core configurations and a 32-core Neural Engine rated at 31.6 trillion operations per second. It supports up to 192 GB of unified memory with 800 GB/s of bandwidth. It shipped in the Mac Studio and Mac Pro (2023).

For AI workloads, the large shared unified memory allows large language models (LLMs) to run locally even when they exceed the capacity of typical consumer GPUs โ€” the CPU, GPU and Neural Engine all draw from the same memory pool without copying data.

#Apple Silicon#M2 Ultra#SoC#Neural Engine#unified memory#UltraFusion#on-device AI
Market classConsumer
Introduction date13 June 2023

Classification

AI System-on-Chip (SoC) ยท serves as: AI acceleration, AI Inference, Compute, High-level compute.

Which group Apple M2 Ultra belongs to and how it is built

5items
Compute
Hardware domain
Consumer
Market class

Integrated System-on-Chip combining CPU, GPU, NPU and memory controllers โ€” used as the primary compute unit in personal computers, mobile devices and robots.

Component type encompassing System-on-Chip designs used as primary compute units in embedded devices, edge AI, and robotics applications.

A single semiconductor component integrating CPU, GPU, AI accelerators, memory controllers, multimedia engines, and communication interfaces.

Technical specification

Neural Engine
2parameters
Neural Engine cores
32
Neural Engine performance
31,6 TOPS

trillion operations per second

Updated: 29 September 2026