小米在玄戒芯片技术沟通会上正式亮相了 AI Cube Prototype 迷你主机工程版1。该产品采用玄戒 O3、玄戒 O100 和玄戒 D100 组成的三芯阵容1，持续性能释放达 150W1，支持 120B/3B 大模型本地部署1。
玄戒 O3 作为 AI 旗舰 SoC，集成 10 核全大核 CPU 架构、16 核 G2-Ultra NX GPU 以及 200 TOPS 低功耗 NPU1。玄戒 O100 采用 6nm3D 晶圆级堆叠先进封装，拥有 28672 根有效数据线，支持 1.22TB/s 超高近存计算带宽1。玄戒 D100 采用 3nm 制程工艺，搭载 20 核高性能 CPU 和 16 核高算力 NPU，最高支持 160GB 内存1。
在工业设计方面，AI Cube Prototype 采用航天铝一体成型工艺，外壳带有 33874 个 CNC 精密镂孔1，支持快慢系统切换1。同时亮相的玄戒 O100 原型机采用横向阔折叠设计2，支持风冷主动散热与 10 瓦强解热能力2，内置 Xiaomi MiMo 端侧模型2。
Xiaomi has officially announced its AI Cube Prototype mini PC, featuring an advanced triple-chip configuration designed to enable local deployment of large language models 12. The device combines the Xuanhe O3 flagship AI SoC, Xuanhe O100 high-bandwidth AI accelerator chip, and Xuanhe D100 autonomous driving AI chip, delivering sustained performance of 150W 1.
The Xuanhe O3 processor features a 10-core all-large CPU architecture and a 16-core G2-Ultra NX GPU paired with a 200 TOPS low-power NPU 1. The Xuanhe O100 employs advanced 6nm 3D wafer-level stacking packaging technology with 28,672 effective data channels, supporting ultra-high near-memory computing bandwidth of 1.22TB/s 1. The Xuanhe D100, built on 3nm process technology, incorporates 20 high-performance CPU cores and 16 high-compute NPU cores, with maximum support for 160GB of memory 1. This tri-chip architecture enables the AI Cube to support local deployment of models ranging from 120B to 3B parameters, with support for quick switching between fast and slow systems 1. The device's enclosure utilizes aerospace-grade aluminum with single-piece forming and features 33,874 precision CNC perforations 1.
At the same event on August 24, Xiaomi also showcased the Xuanhe O100 prototype device, which employs a horizontal wide-folding design and features active wind cooling with 10-watt heat dissipation capability 2. The prototype integrates the Xiaomi MiMo edge-side model, enabling exceptionally high edge inference speeds 2.
评论
还没有评论，欢迎留下第一条。