Apple M5 Silicon: Major Redesign with Split CPU and GPU Memory
Rumours are swirling about Apple’s next-generation M5 silicon, and they hint at a significant redesign. Unlike its predecessors, the M5 lineup may abandon the unified memory architecture that has been a hallmark of Apple silicon. Instead, Apple could implement separate memory pools for the CPU and GPU, marking a bold departure from its established approach.
Unified Memory: A Legacy Redefined
Since the debut of the M1 chip, Apple silicon has utilized Unified Memory Access (UMA). This design, which shares memory between CPU and GPU cores, has been credited with delivering exceptional performance-per-watt efficiency in devices like MacBooks and Mac Studios. However, industry analyst Ming-Chi Kuo reports that the upcoming M5 chips will pivot to a split memory design.
While separating CPU and GPU memory adds complexity, it could yield significant performance advantages for specific workloads. This change aligns with Apple’s ambition to push the boundaries of silicon design.
Advanced Packaging with TSMC’s SoIC-mH
To support this architectural shift, Apple is reportedly leveraging TSMC’s advanced 2.5D packaging technology, known as System on Integrated Chips-Molding Horizontal (SoIC-mH). Unlike vertical 3D stacking, this approach integrates separate dies horizontally within the same package, enabling high-density connections while improving yields and thermal management.
This innovation is expected to unlock new levels of performance, particularly in demanding applications like AI processing. Apple’s Private Cloud Compute, the company’s AI-focused cloud intelligence system, stands to benefit significantly from these advancements.
Fabrication on TSMC’s 3nm N3P Node
The M5 lineup will be fabricated using TSMC’s N3P process node, an improvement over the N3E node used in current-generation M4 and A18 Bionic chips. The N3P node offers higher transistor density and energy efficiency, reinforcing Apple’s leadership in chip performance.
Timeline for Release
According to Kuo, the base M5 chip will enter mass production in early 2025, with devices hitting the market later that year. The M5 Pro and M5 Max are expected in late 2025, while the flagship M5 Ultra could debut in 2026.
A Revolutionary Step for Apple Silicon
The rumoured split memory design signals a transformative phase for Apple silicon. By leveraging TSMC’s cutting-edge technology, the M5 lineup promises to deliver groundbreaking performance, paving the way for more powerful Macs and AI-focused applications. Apple is once again pushing the boundaries of innovation in silicon design.
