Baidu Intelligent Cloud releases the energy-efficient cloud server GR1 instance based on Ampere® Altra®
In 2021, Baidu announced its 2030 carbon neutrality goal and 6 scientific paths to achieve it. The first step is to build greener data centers. Baidu's self-built data center PUE has always been in a very leading position. Baidu has always adhered to the sustainable development concept of "technology for better", relying on full-stack and leading artificial intelligence technology and rich industrial practices. On the basis of achieving its own "carbon neutrality" goal, it will also use technology output to serve the whole world. Social energy conservation and emission reduction assistance.
In the process of digital transformation, diversification and heterogeneity of computing power have become a development trend. In order to solve the challenges of growing demand for computing power and the trend of energy conservation and emission reduction, in addition to technical upgrades of existing data center infrastructure, more efficient solutions can also be adopted without increasing data center space and energy consumption. , to achieve synergy between low-carbon development and computing power supply.
To this end, Baidu Smart Cloud officially released the new BCC instance product Gr1. Different from other cloud servers, the Ampere ® Altra ® cloud-native processor equipped on Gr1 is based on the Arm instruction set architecture and adopts a single-core single-thread design, allowing each computing core of the cloud instance to exclusively enjoy ALU (logical computing unit), cache, etc. Key physical resources, each core has a constant main frequency of 3.0GHz, achieving stable and predictable performance and better security. More importantly, when providing the same performance, the power consumed by Gr1 instances will be reduced by more than half. Not only can every Baidu Smart Cloud user enjoy the new features brought by Gr1, but it can also achieve active emission reduction.
Baidu Smart Cloud Gr1 cloud instance supports core-to-storage ratios of 1:2 and 1:4, with optional core numbers from 1 core to 64 cores, supports ESSD cloud disks, supports up to 25Gbps intranet bandwidth, and has ultra-high network packet sending and receiving capabilities. , which can meet extremely high intranet transmission requirements.
Gr1 instance products are ideal for the most common cloud-native application workloads, such as open source databases, Java applications, video encoding and decoding, AI inference (data governance), big data analysis, DevOps, web service applications, etc. Compared with traditional cloud servers, Gr1's performance shows obvious advantages. For example, for three common workloads in Web service applications: NGINX, Redis and MySQL, Gr1 can achieve exponential price-performance improvements.
Today, AI inference has been widely used in digital applications. For large-scale AI models or high-performance scenarios, most of them are implemented using GPUs or dedicated chips. However, for many independent data service systems such as Web applications or data management, using GPU platforms is not the most cost-effective choice. As its workload fluctuates dynamically, the workload of the GPU is rarely saturated, and the computing power of the GPU is not always fully utilized, making it difficult to achieve flexible expansion as the actual business volume changes.
At this time, it is most common to use CPU for AI inference, and standard cloud servers with rich configurations can meet business needs. In addition to lower cloud server acquisition costs, users will have better control, no reliance on third-party hardware, lower software complexity involving operating systems, drivers, runtimes, and easier scalability and integration with other software stacks Integrate with microservices. Baidu Smart Cloud Gr1 instances support optimized AI frameworks (TensorFlow, PyTorch, ONNX), which can be used out of the box and can be seamlessly integrated without any modification. The Ampere Altra processor used in Gr1 natively supports the fp16 data format, unleashing ultimate performance without loss of precision.
In recent years, there have been emerging Arm native applications originating from the terminal market, such as cloud terminal emulation, cloud mobile phones, cloud games, large-scale App development and testing, etc. Thanks to the isomorphism of the instruction set, Gr1 instances natively support 64 & 32-bit applications. It is perfectly applicable. Using cloud Gr1 instances, you can make full use of efficient cloud development platforms and tool chains to achieve cross-device portability and cross-platform compatibility from cloud/edge to device.
Currently, Gr1 instances already support common Linux public images such as CentOS7.9, CentOS7.6, Ubuntu22.04, and Ubuntu20.04. The next generation of computing technology will be designed with cloud-native software technologies (such as microservices and containers). In order to make it easier for customers to deploy cloud-native workloads, Cloud Container Engine (CCE), the key infrastructure of Baidu Smart Cloud containerized workloads, currently provides support for Gr1 instances, and more cloud products will be enabled in the future. .
Baidu Smart Cloud has many years of experience in Arm architecture technology and will continue to participate in the increasingly complete Arm ecosystem to help cloud users build excellent solutions on the more efficient Gr1 platform and gain better cost performance. At the same time, through Baidu Intelligent Cloud technology, more small and medium-sized enterprises and software developers can participate in carbon emission reduction.
Source of pictures and text in this article: Anshengpei Semiconductor
Click here to read the original article
京公网安备 11010802033920号