计算机行业：Meta发布Sapiens视觉模型，落地场景广阔

Industry Investment Rating - The report gives a "Recommend" rating for the computer industry [1] Core Views - Meta's Sapiens AI visual model marks a new era in human motion intelligence analysis, excelling in four human-centric visual tasks: 2D pose estimation, body part segmentation, depth estimation, and surface normal prediction [1] - The Sapiens model, with parameters ranging from 300 million to 2 billion, uses a visual transformer architecture and natively supports 1K high-resolution inference [1] - Potential applications of the Sapiens model include AR/VR, 3D human digitization and animation, human-computer interaction systems, autonomous driving, and intelligent surveillance [1] Global Market Performance Stock Market Dynamics - US and Chinese tech stocks experienced a pullback, while Hong Kong tech and A-share computer stocks rose [8] - Key index changes: S&P 500 +0.24%, Nasdaq Composite -0.92%, Philadelphia Semiconductor Index -1.34%, TAMAMA Tech Index -1.70%, Nasdaq Golden Dragon China Index -0.22%, Hang Seng Tech Index +2.64%, Computer Index +2.60% [8] Bond and Forex Markets - US 10-year Treasury yield rose to 3.91%, up 10bps [10] - USD/CNY central parity rate was 7.11, up 234 basis points from August 23 [10] Key Tech Stock Performance - Apple +0.95%, Nvidia -7.73%, Tesla -2.82%, Google -1.35%, Amazon +0.82%, Meta -1.27%, Microsoft +0.08%, ARM -2.03%, Intel +7.30%, Qualcomm +1.04%, AMD -4.14% [14] Industry News Computing Power and Terminals - Microsoft unveiled the Maia 100 AI chip, designed for large-scale AI applications in Azure cloud services, featuring 105 billion transistors and 820mm² chip size [16] - IBM introduced the Telum II processor and Spyre accelerator, enhancing AI capabilities for IBM Z mainframe systems [17] - Intel announced Granite Rapids-D Xeon 6 SoC, expected to launch in H1 2025, with Intel 3 and Intel 4 process technologies [18] Large Models and Cloud Applications - IBM Cloud will be the first to deploy Intel Gaudi 3 AI accelerators, expected in early 2025 [19] - Google's GameNGen, a neural network-driven game engine, can generate real-time game frames without traditional game engine components [20] - Meta's Sapiens AI visual model excels in human motion analysis, with applications in AR/VR, 3D animation, and autonomous driving [21]