Tencent Cloud launched the first high-performance heterogeneous computing infrastructure in China, the FPGA cloud server, to promote the FPGAs that large companies can use for long-term payment to more enterprises in the form of cloud services. With the FPGA cloud server, enterprises can perform FPGA hardware programming, which can increase the performance to more than 30 times that of a general-purpose CPU server, and only pay about 40% of the cost of a general-purpose CPU. Tencent Cloud became the first cloud service provider in China to open FPGA computing services in the cloud. FPGAs have hardware-programmable, low-power, low-latency features that represent the future of high-performance computing compared to the already-represented high-performance computing GPU. Enterprises can use FPGAs in the deep learning phase of the field, and use FPGAs in the detection phase of deep learning, complementing the GPUs mainly used in the training phase. Not only that, FPGAs can also be used in areas such as financial analysis, image and video processing, genomics and other areas where high performance computing is required, and are the best choice for such industry applications with high efficiency requirements. For example, the financial industry processes transaction data in real time, using a general-purpose CPU to handle large delays and high costs, and using FPGA can solve the above problems well. FPGA is also deeply applied by Internet search companies in the search field because of low latency. The reason why FPGAs are more efficient than CPUs and even GPUs is essentially the benefits of an architecture without instructions and without shared memory. Wang Huixing, vice president of Tencent Cloud, believes that FPGAs will change the design logic of artificial intelligence from the bottom, accelerate the processing capability of enterprise business in the cloud, and promote the innovation pace of the whole industry in the field of artificial intelligence. He further pointed out that FPGA has been used for a long time in Tencent's massive business such as QQ and WeChat. It has deep technical accumulation and complex scene practice. It is hoped that this technology opening will greatly reduce the threshold for enterprises to use, so that large and medium-sized enterprises can catch up. Artificial intelligence train. FPGA has been in existence for more than 30 years. It has been mainly used in large-scale Internet and technology companies. Tencent Internet companies such as Tencent have deployed FPGAs in batches in data centers to provide hardware acceleration platforms for their deep learning applications. Compared with traditional computing architectures, FPGAs can better support large-scale parallel computing requirements such as deep learning. Compared with the general-purpose CPU horizontal test, FPGA shows obvious advantages in cost and performance in application scenarios such as deep learning, image compression and classification detection. FPGA is widely used in Tencent's deep learning field. Deep learning is mainly used in image classification and recognition, speech recognition, and personalized content recommendation algorithms. Among them, WeChat and QQ in order to enhance the processing ability of the illegal picture detection of social communication and other services, reduce the cost of picture detection, and accelerate the CNN algorithm of the deep learning model through FPGA. The processing performance of the FPGA is 4 times that of the general-purpose CPU, and the unit cost is 1/3 of the general purpose CPU. The advantage of FPGA in image processing is also obvious. Users upload huge amounts of pictures every day through social platforms such as WeChat and QQ. If WEBP image compression format is used instead of standard JPEG format, the picture size can be reduced by one-third, and the user can bring Come to a better picture download experience. Accelerated testing of the image transcoding of 853x640 by FPGA shows that the processing delay of FPGA is 20 times lower than that of general-purpose CPU, the processing performance is 6 times that of general-purpose CPU, and the unit cost of FPGA is only 1/3 of general-purpose CPU. . Another test shows that FPGAs have obvious advantages in DNN calculation acceleration. Domestic and foreign search engine companies generally use the DNN algorithm for ad search. Tencent uses FPGA to accelerate the DNN calculation. It uses the 50% FPGA resource to reduce the DNN calculation delay of 4000 samples to 1.2 milliseconds, the throughput rate reaches 6000 sets / s (4000 samples for 1 set), not only meets the system's low latency requirements, but also greatly improves the system's throughput: processing latency is reduced by 100 times, processing throughput Raised by 5 times, and the cost is 1/5 of the general purpose CPU server. In the past, SMBs faced many problems in deploying FPGAs: FPGA hardware costs are high, flexibility is poor, and one-time procurement is expensive. It requires high cost and great effort for FPGA customization and procurement, and requires hardware engineers and software engineers. At the same time, enterprises will also face the problem of resource idle circulation caused by FPGA chip replacement. Although FPGA Intellectual Property (FPGA IP) provides the hardware acceleration required for the business, the long development cycle, high R&D investment and high risk make many companies discouraged. The revolutionary development brought by Tencent Cloud is to reduce the FPGA deployment time from several months to several minutes. Enterprises can use FPGAs on demand, which greatly reduces the cost of FPGAs and accelerates the processing of high-performance FPGA hardware. At the same time, Tencent Cloud is the first to provide third-party FPGA intellectual property market in China. Through the Tencent cloud service market, FPGA developers and users can trade more efficiently. For a long time, providers and users in the FPGA industry have been lacking in trading platforms and credit guarantees, resulting in lengthy transactions and difficulty in reaching deals. Tencent cloud service market builds a simple and reliable FPGA intellectual property trading market. For FPGA users, you can purchase developed and validated FPGA intellectual property functions in the Tencent cloud services market, saving up to several months of FPGA development cycles, while using pay-as-you-go models to maximize savings. Hardware input costs. Wang Huixing believes that for the cloud computing industry, providing FPGAs in the cloud is a revolutionary advancement in cloud computing IaaS. From the bottom, it accelerates the application of cloud computing in various scenarios. Cloud computing will move from pure resource services to artificial intelligence clouds. Service era. Smart Bms,Bms For Battery,Bms For Lithium Battery,Bms Module HuiZhou Superpower Technology Co.,Ltd. , https://www.spchargers.com
FPGA developers can significantly improve R&D efficiency by using the development framework of Tencent Cloud FPGA, and focus on core function development. The existing mature FPGA intellectual property such as image processing function and deep learning function can be integrated and integrated by simple package. It is put into the Tencent cloud service market and is open to FPGA users to share the research and development costs of FPGA intellectual property.