微软近日宣布，将在明年初为 DirectML 机器学习框架添加 NPU 支持，从而适配英特尔酷睿 Ultra 等内置 NPU 的芯片。这一举措将极大地改善 AI 模型运行效能，使得开发者能够更好地利用芯片内置的硬件资源。
微软在公告中表示，DirectML 是一款基于微软自家 Windows 平台的开源机器学习框架，它已经在许多领域得到了广泛的应用，如图像识别、语音识别等。而英特尔酷睿 Ultra 等芯片内置的 NPU 则是一种专门为深度学习加速设计的硬件单元，它可以大大提高 AI 模型的计算效率。
通过这次升级，DirectML 将支持 NPU，并提供跨平台推理引擎 ONNX Runtime 的 API 接口，这使得开发者可以方便地使用芯片内置的 NPU 运行 AI 模型，大大提升了模型的运行效率和性能。此外，DirectML 还将支持更多的神经网络架构和优化技术，为开发者提供更多选择。
对于人工智能行业来说，这次升级无疑是一个重要的里程碑。随着越来越多的芯片厂商加入到 AI 领域的竞争中来，我们有理由相信，未来的 AI 技术将会更加成熟和普及。
英语如下：
News Title: “Microsoft DirectML to Support NPUs, AI Model Performance May Improve”
Keywords: Microsoft, DirectML, NPU
News Content: Microsoft recently announced that it will add support for NPUs to its DirectML machine learning framework in early next year, making it compatible with chips such as Intel Core Ultra that have built-in NPUs. This move is expected to greatly improve the performance of AI models, allowing developers to better utilize the hardware resources built into the chip.
In the announcement, Microsoft stated that DirectML is an open-source machine learning framework based on its own Windows platform, which has been widely used in various fields such as image recognition and speech recognition. The NPU built into chips like Intel Core Ultra is a hardware unit specifically designed for accelerating deep learning, which can significantly increase the computing efficiency of AI models.
With this upgrade, DirectML will support NPUs and provide an API interface for the cross-platform inference engine ONNX Runtime, making it easy for developers to use the built-in NPU to run AI models, greatly improving model running speed and performance. In addition, DirectML will also support more neural network architectures and optimization technologies, providing developers with more options.
For the artificial intelligence industry, this upgrade is undoubtedly an important milestone. With more and more chip manufacturers joining the competition in the AI field, we have reason to believe that future AI technology will become more mature and popular.
【来源】https://www.ithome.com/0/740/014.htm
Views: 1