MediaTek Unveils Dimensity 9600 Pro With 2nm Process and AI-Native Fusion Computing Architecture
MediaTek has launched the Dimensity 9600 Pro, its first Pro flagship mobile platform, featuring TSMC's 2nm process, an AI-native fusion computing architecture, dual NPUs, a 2+3+3 all-big-core CPU, neural rendering and upgraded imaging.
The report says the Dimensity 9600 Pro is among the industry's first chips to use TSMC's 2nm process and integrates more than 33 billion transistors. MediaTek worked with TSMC on 2nm development and used design-technology co-optimization to push the process's performance ceiling. Its Dimensity fusion computing architecture links CPU, NPU, memory and storage, and is described as an agent-oriented CPU architecture with scheduling technology intended to improve performance and energy efficiency. The report says ultra-large core power consumption fell 37 percent and multi-core power consumption fell 61 percent.
The chip adopts a 2+3+3 all-big-core CPU design with two C2-Ultra cores at up to 4.55GHz, three C2-Pro cores at 4.35GHz and three C2-Pro cores at 3.1GHz. It includes a 34.5MB L2+L3+system cache, with L2 capacity up 21 percent over the previous generation, and supports LPDDR6 memory and UFS 5.0 storage. In Geekbench V6.4, the report cites a single-core score of 4,276, up 17 percent, and a multi-core score of 12,650, up 15 percent. AI scheduling predicts task demand and dynamically allocates system resources; for a 3B large language model, answer token output speed rises by up to 40 percent. AI model launch speed improves by 27 percent, and the platform supports 25 apps running concurrently while keeping them alive.
For AI processing, the Dimensity 9600 Pro uses a dual-NPU setup: the ultra-performance NPU 1090 and a second-generation ultra-energy-efficient NPU. The energy-efficient NPU is designed for always-on AI tasks and cuts power consumption by 40 percent compared with the previous generation, allowing continuous sensing of sound, motion and environmental information. NPU 1090 increases on-device AI inference, with tokens generated per watt up 55 percent. The platform supports on-device 30B MoE large models, upgrades the Dimensity AI agent engine to version 3.0, adds the Sensing Foundation suite for environment-aware scheduling, supports Hybrid Attention computing and on-device MoE models, and includes the AISeal on-device AI privacy protection architecture covering data isolation, verification tracing and unified permission management.
The report describes several partner experiences built on these capabilities. A camera assistant developed with vivo uses the energy-efficient NPU to sense the viewfinder, recognize scenes and suggest camera functions. An Inspiration Composition feature with OPPO analyzes subjects, elements and their relationships before offering composition advice. vivo's AI One-Tap Highlight runs a multimodal model on device to identify expressions and emotional changes in videos and automatically create highlight clips. An upgraded Xiaobu assistant with OPPO uses chip-level capabilities to gather information across apps and remind users about meetings and trips. MediaTek is also exploring active AI experiences with StepFun, where a phone can judge content based on habits and real-time context and, after confirmation, help plan trips such as flights and hotels after seeing concert information.
The GPU is the G2-Ultra NX, with a third-generation ray tracing engine, Raytracing Pipeline technology and hardware OMM ray tracing. MediaTek says GPU performance is up 18 percent over the previous generation. The larger change is the Dimensity neural rendering architecture, which lets AI compute units participate in graphics rendering alongside the GPU. In Honor of Kings, native 720p CG footage can be AI real-time super-resolved to 1080p; in Where Winds Meet, native 720p can be raised to 1080p. In Genshin Impact at extreme 864p, same-frame power consumption falls 10 percent, and in a leading anime turn-based mobile game at 60FPS, same-frame power falls 15 percent. Dimensity frame generation 4.0 reaches up to 185FPS, with per-frame power 21 percent lower than native 90FPS.
For imaging, the Dimensity 9600 Pro introduces a new intelligent imaging architecture in which ISP and NPU work together. The report says it debuts mirrorless-level full-depth computational optical imaging, using NPU multi-frame composition to keep foreground and background clear in group photos. For action shots, the Dimensity rapid focus engine with NPU AI raises professional sports-level tracking to 60FPS, with tracking and focus speeds both twice as fast as the previous generation. It supports 100dB dynamic light capture and 17EV high-dynamic imaging, an 8x increase in light contrast capture. Video features include 4K240 slow motion, 4K120 cinematic Log native recording, an independent hardware-level original-color recording engine, CSISP 4K60 cinematic video and multi-camera asynchronous parallel processing.
Display and connectivity also receive upgrades. The chip debuts full-link BT.2020 support and H.266 hardware decoding, and MediaTek says it is the first on Android to support multi-frame OverDrive panel response compensation, reducing smear rate by 30 percent compared with single-frame OverDrive. The Dimensity AI communication open platform can run AI network models inside the communication module with 50 percent lower power than running models on the CPU. AI cellular network selection 2.0 reduces lag duration by 30 percent, with a further 40 percent reduction after coordination with OEM large models. AI positioning 2.0 improves positioning accuracy by 30 percent.