联发科发布天玑 9600 Pro 旗舰芯片：2nm 工艺、120 帧光追游戏

联发科于 9 月 15 日正式发布天玑 9600 Pro 旗舰芯片 1。该芯片采用行业首批 2nm 制程工艺，集成 330 亿余枚晶体管，由台积电共同研发 1。

在性能方面，天玑 9600 Pro 搭载 2 个超大核心运频达 4.55GHz，配合全新 Pro 架构设计，相比前代天玑 9500，CPU 单核性能提升 17%、多核性能提升 15% 1。GPU 同功耗下性能提升 18%，峰值性能工作时功耗下降 24%；光追性能亦提升 18%，支持手机游戏 120 帧光追运行 1。

AI 能力方面，该芯片 AI 问答出词速度提升 40%，每瓦特生成词元数提升 55%，大语言模型预填充性能提升 51% 1。能效方面，在线视频播放功耗下降 48%，NPU 轻载模型功耗下降 40% 1。此外，L2 缓存提升 21%，系统缓存达 34.5MB 1。

视频录制方面，天玑 9600 Pro 支持 4K 240 帧超高速慢动作录影以及 4K 120 帧电影级 Log 原色视频录制 1。定位精度相比前代产品提升 30% 1。

MediaTek officially unveiled the Dimensity 9600 Pro flagship processor on September 15, 2024, marking the company's latest advancement in mobile chip technology 1. The processor is built on a 2nm manufacturing process—among the first in the industry—and contains over 33 billion transistors, developed in collaboration with Taiwan Semiconductor Manufacturing Company 1.

The Dimensity 9600 Pro features a configuration of two ultra-large cores clocked at 4.55GHz, three large cores at 4.35GHz, and three additional cores at 3.1GHz 1. The new Pro architecture delivers a 17% improvement in single-core CPU performance and a 15% boost in multi-core performance compared to the Dimensity 9500 1. The chip supports ray-traced mobile games at 120 frames per second and enables 4K ultra-high-speed slow-motion recording at 240 frames as well as 4K 120-frame cinema-grade log color video capture 1.

Beyond gaming and video capabilities, the processor demonstrates significant gains in artificial intelligence and power efficiency. L2 cache increases by 21%, with system cache capacity reaching 34.5MB 1. AI response word generation speed improves by 40% 1. The chip achieves a 40% reduction in NPU light-load model power consumption and a 55% improvement in token generation per watt for large language models, alongside a 51% enhancement in large language model prefill performance 1. GPU performance shows an 18% improvement in ray-tracing performance and an 18% increase in peak performance under equivalent power consumption, while peak performance power consumption decreases by 24% 1. Additional features include a 30% improvement in positioning accuracy and a 48% reduction in online video power consumption 1.