Table of contents
embrace artificial intelligence
Extreme performance optimization
foreword
At present, China's source code contribution has reached the second place in the world, and the number of open source software developers has also exceeded 8 million, ranking second in the world. With the attention of many developers, the 2023 Open Atom Global Open Source Summit was grandly held in Beijing. All major open source leading companies have come up with their own "housekeeping skills", such as Alibaba's Dragon Lizard Community, Huawei Cloud's CCE Turbo, Tencent Cloud's TDSQL, Baidu's XuperChain... But what attracts me the most is Intel's booth and an open source forum for collaboration between hardware and software.
Intel's Open Source Path
As the absolute leader in the processor field, Intel open-sourced the source code of the 6-bit microprocessor 8086 as early as the 1980s, thus opening the glorious era of the x86 architecture. The personal computer based on the x86 architecture has led the Internet boom for more than 20 years. This open source has greatly promoted the development of processor technology, and it is still used as a typical textbook by computer scientists.
In more than 40 years, Intel has never stopped on open source. Intel's contribution can be seen in many prestigious open source projects, such as Linux kernel, Tensorflow, TBB, MKL, Kubernetes, etc. So far, Intel has been the largest contributor to the Linux kernel, led more than 100 open source projects, participated in the maintenance of more than 300 community projects, and is also a member of more than 700 standard organizations.
In the open source forum for software and hardware collaboration, Intel scientists and engineers shared Intel's software open source ecological construction, cloud-native open source cloud solutions, open source cross-platform Android tool Celadon, etc. However, as an algorithm engineer, what attracted me the most was Intel's advantages and innovations in artificial intelligence shared by Dr. Zhang Yu, Intel's senior chief AI engineer.
embrace artificial intelligence
The crest of AIGC's wave
The peak of the wave of artificial intelligence has arrived. This is a new era, and intelligence is becoming a reality. With the development of artificial intelligence technology, intelligent applications are making people's lives more convenient and efficient. In the past few decades, scientists have been exploring artificial intelligence technology and achieved many major breakthroughs. These breakthroughs include technologies in machine learning, deep learning, natural language processing, and computer vision, which are being widely used in production, service, medical care, education, and other fields.
In recent years, the large language model (LLM) represented by ChatGPT and the multimodal generation model represented by Stable Diffusion are deeply affecting us. At the summit booth, Intel engineers enthusiastically introduced Intel's support for artificial intelligence, especially for generative models, and demonstrated the accelerated deployment of stable diffusion, which can generate a high-resolution image in a few seconds. The stunning images and stunning effects won unanimous praise from the developers present.
At the peak of the AIGC wave, many excellent artificial intelligence companies have emerged. Through their own R&D and innovation, these enterprises continue to promote the development of artificial intelligence technology and create more artificial intelligence applications and commercial value. At the same time, these companies are also actively participating in the artificial intelligence open source community to promote the opening and sharing of artificial intelligence technology. Intel is undoubtedly one of the companies that plays an important role in the artificial intelligence open source community. Intel has been committed to the research and innovation of artificial intelligence technology, and has launched many excellent artificial intelligence products and solutions, especially the full-link software and hardware. Collaborative artificial intelligence solutions, combining Intel's advantages in hardware and software, can provide a full range of artificial intelligence solutions, from algorithm design to computing power support, from application development to system debugging, and realize the whole chain of artificial intelligence technology road support.
Full Link AI Solution
Intel's full-link software-hardware collaborative artificial intelligence solution can provide more efficient and optimized computing power support, making full use of Intel's years of accumulation in hardware, providing efficient parallel computing and optimized data processing capabilities, especially Intel's launch of Processors optimized for AI at the edge: Core™ processors, Data Center GPU Flex series, Xeon® Scalable processors, and more. These processors and chips are optimized for different edge artificial intelligence scenarios, featuring high efficiency, low power consumption, and high performance. For example, with the support of the Media Computing Engine and Xe Matrix Extensions (XMX) on the Data Center GPU Flex processor, the performance of the Flex 170 in various combinations of encoding formats and AI models has surpassed that of mainstream solution providers in the industry. . Especially in artificial intelligence visual reasoning, it can be used in combination with multiple scenarios. By decoding and processing video streams and then performing AI calculations, it realizes the application scenarios of multiple workload integration and demonstrates strong flexibility.
In addition to these AI-optimized hardware, Intel's full-link software-hardware collaborative AI solution also provides very rich and complete software support. For example, well-known deep learning frameworks such as TensorFlow, CNTK, and Caffe all have contributions from Intel. In addition, Intel provides a variety of artificial intelligence development tool kits, including Intel® Extension for Pytorch, Intel® Extension for Tensorflow, Intel® Advanced Matrix Extension Engine, and Intel® Nervana™ Toolkit, etc. Performance, software and hardware co-optimization to accelerate computing. These tool suites cover the whole process of deep learning model training and deployment inference optimization, and can help developers quickly build their own artificial intelligence applications.
Extreme performance optimization
Although many manufacturers currently provide artificial intelligence solutions, Intel's software-hardware collaborative full-link solution can release performance to the greatest extent and achieve the most extreme optimization. At the scene, Dr. Zhang Yu focused on Intel's open source suite OpenVINO, which is designed based on convolutional neural network (CNN) and supports deep learning reasoning from the edge to the cloud. OpenVINO can realize "write once, deploy anywhere", convert the compiled model into OpenVINO common format, and can expand the work on one-click deployment on Intel hardware platforms (including accelerators) including Intel CPU, GPU, VPU and FPGA load and maximize performance.
In the latest version release, OpenVINO supports generative AI models, such as GPT, CLIP, BLIP, Stable Diffusion 2.0, ControlNet, etc., and has made many extreme optimizations for these models and underlying operators. A very comprehensive performance optimization guide is also provided, including an accuracy checker, post-training optimization tools, and other tools for accuracy measurement, performance benchmarking, and application tuning. The OpenVINO development kit allows more developers to quickly get started in the wave of AIGC, not only to quickly deploy their own AI tasks, but also to maximize inference performance.
conclusion
The era of AI is approaching quietly, and the development of science and technology is changing the lives of each of us everywhere. Intel's continuous efforts in open source in China will jointly contribute to the ecological prosperity of China's open source technology. In this best era, let us embrace AI and enjoy the top of this wave.
