On December 15, Intel announced at the official launch event of the first generation Core Ultra that Alibaba Cloud's Tongyi Qianwen model was successfully adapted to the Intel Core Ultra platform. This means that in the future, this powerful large language model can be deployed on the PC and other devices. Thanks to the discrete module architecture of the Intel Core Ultra platform (codenamed Meteor Lake) and the first integrated neural network processing unit (NPU), the Tongyi Qianwen large model deployed on the Intel Core Ultra platform device can significantly reduce the model inference response time, with an average acceleration of up to 3x.
Previously, Intel also tested the good performance of the LLAMA2-7B large model from Meta on the device side without a network, which made the concept of AI PC gradually clear. Since the release of ChatGPT at the end of last year, AIGC applications supported by large language models have developed greatly, and the main position of this wave of AI has occurred on PC, because PC is more suitable for most AIGC applications in terms of interaction and presentation. But whether it's ChatGPT or Midjourney, you need to be connected to the Internet, and once there is no network, then we will be disconnected from the AI.
Therefore, if a PC wants to truly evolve into an AI PC, not only can the AIGC application be easily run on it, but it also needs to have strong AI processing power. The news that Alibaba Cloud Tongyi Qianwen large model QWEN-7B has been successfully adapted to the Intel Core Ultra platform and can be deployed on PC and other devices in the future has undoubtedly brought new possibilities to the development of PC.
At the ON Technology Innovation Conference held in Silicon Valley in September this year, Intel proposed a revolutionary concept of PC - AI PC. Intel CEO Pat Gelsinger said that if you want to efficiently perform AI computing and run AI applications on PCs, the traditional CPU+GPU architecture is not the optimal solution. As a result, on the first-generation Core Ultra platform, Intel innovatively adopted a discrete module architecture, which is also the largest processor structure change in Intel's history. The processor will consist of four separate modules: the compute module, the SOC module, the graphics module, and the IO module, and will be connected via FoverOS 3D packaging technology.
Among them, the new NPU module will mainly assume the responsibility of AI computing, because it is a special computing unit for artificial intelligence acceleration, and the advantage of AI computing in the NPU model is high performance and low power consumption. In addition, low-latency and high-response CPUs and high-performance and high-throughput GPUs will also afforded the demand for AI computing power, and the combination of the three will form a powerful AI computing power on the device. Intel said that the overall AI computing power of the Core Ultra 7 165H not only beats its predecessor Core i7-1370P, but also crushes its rival AMD 7840U.
In the case of the Core Ultra 7 165H processor (TDP 28W), Intel says that its multi-threaded performance is more than 3x that of the Qualcomm Snapdragon 8CX Gen 3 and 1 of the direct competitor AMD 7840U11 times. At the same time, Intel said that the Core Ultra 7 165H also outperforms its rival AMD 7840U in terms of energy efficiency in various scenarios with better multi-threaded performance. In addition, the low-power island is a very interesting design, with 2 LPE-cores ultra-low-power cores, which can run tasks that require less responsiveness, such as *** and so on, further reducing power consumption.
Now we can perform a variety of AI operations on PC, including but not limited to Wensheng Wen, Wensheng Diagram, Wensheng**, as well as AI cutout, gesture recognition, motion capture, etc. Some of them are new experiences based on large language models, and some are already available on PC. For example, a large model like Tongyi Qianwen already has certain general capabilities and can provide developers from all walks of life with a variety of choices**.
The X-D Lab team of East China University of Science and Technology has developed a number of industry models based on the Qwen-7B and QWEN-14B open-source models of Tongyi Qianwen, including the mental health model MindChat, the medical and health model Sunsimiao, and the education examination model GradChat (Koi). Based on QWEN-7B, Zhejiang University and Higher Education Press have developed a vertical model of Zhihai-Sanle education, which has been applied in 12 universities across the country. Zhejiang Youlu Robotics*** has integrated QWEN-7B in the pavement cleaning robot, enabling the robot to interact with users in real time in natural language.
2023 Annual Review In short, the stage of AI is not only in the PC, and the PC is not only in AI applications, but there is a huge intersection between them. Every user who owns an AI PC will gradually face changes in their choice and Xi habits. The concept of AI PC is gradually becoming clear, and if PC wants to truly evolve into AI PC, not only can AIGC applications be easily run on it, but it also needs to have strong AI processing capabilities. The news that Alibaba Cloud Tongyi Qianwen large model QWEN-7B has been successfully adapted to the Intel Core Ultra platform and can be deployed on PC and other devices in the future has undoubtedly brought new possibilities to the development of PC.