Apple to Kick Off M5 Chip Development in H1 2025 With Separate CPU-GPU Design: Ming-Chi Kuo

Apple will kick off the mass production of its M5 family of chipsets next year, according to claims by a supply chain analyst. Its volume production is tipped to take place leveraging TSMC's 3nm semiconductor node (N3) which may offer performance gains and increased efficiency compared to the previous manufacturing process. Apple is also speculated to employ a new chipset design which could take up less space than conventional system-on-chip (SoC) architecture.

Apple's M5 Chip Development

According to Ming-Chi Kuo, a supply chain analyst at TF International Securities, the standard M5 Apple silicon chip will enter mass production in the first half of 2025. Its development is tipped to be followed by its other family members such as the M5 Pro/Max and Ultra variants that may enter the fabrication stage in 2H25 and 2026, respectively.

The M5 chipset will be fabricated using TSMC's N3P node which could deliver a 5 percent performance gain while also reducing the power draw by 5-10 percent compared to M4 SoC's N3E node, as per Kuo. The Cupertino-based technology giant is expected to employ a “server-grade” System-on-Integrated-Chips-molding-Horizontal (SoIC-mH) design with its M5 Pro and Ultra chips that would have multiple integrated circuits integrated into a single unit, with chips being placed horizontally within the mould.

The analyst suggests that this unconventional design could take up 30-50 percent less space than conventional system-on-chip (SoC). This method is tipped to result in improved thermal performance and less throttling.

Further, the iPhone maker will also separate its CPU and GPU design on the M5 chip. As per Kuo, this separation could result in even more performance gains. Apple currently integrates both CPU and GPU on the same chip. If they are separated, each component would be optimised for its specific role, in theory. Kuo suggests that M5 SoC might also deliver a boost when it comes to artificial intelligence (AI) tasks.