Apple unveils M6 and M5 Ultra chips

Summary

Apple has announced the M6 chip (its first 2 nm Apple silicon, featuring a 12-core CPU, dual 16-core Neural Engine, and 170 GB/s bandwidth) and the M5 Ultra (the first four-die Ultra design with an UltraFusion interconnect operating at 4.4+ TB/s, supporting up to 36 CPU cores, 80 GPU cores, and 512 GB of unified memory). Both chips are positioned to run large AI models locally on Macs.

Why it matters

The 2 nm manufacturing process and four-die UltraFusion architecture signal a structural shift in Apple’s silicon scaling, as the company moves beyond single-die limitations. With 512 GB of on-device memory and 1.2 TB/s bandwidth, the M5 Ultra targets workloads that previously required cloud services or discrete GPU setups, including large language models (LLMs) with hundreds of billions of parameters, 3D rendering, and scientific computing. This move strengthens Apple’s competitive position against NVIDIA and AMD in the professional content creation and local AI inference markets.

Highlights

• M6 is Apple’s first 2 nm chip; its dual 16-core Neural Engine doubles AI computing power compared to the previous generation.
• M5 Ultra is the first four-die Apple Silicon to utilize the new UltraFusion interconnect operating at 4.4+ TB/s.
• 512 GB of unified memory and 1.2 TB/s bandwidth are offered to enable fully on-device execution of LLMs with hundreds of billions of parameters.

Critical take

All performance data in the article relies on Apple’s own marketing claims (peak/maximum efficiency, best-case comparisons). The source is a regional tech portal merely republishing a press release. No independent benchmarks, thermal data, or efficiency/cost context are provided. The modest 1.2x single-thread improvement of the M6 over the M5 and the complete absence of real-world workload tests make it difficult to distinguish genuine architectural progress from selective benchmarking.