Shulou(Shulou.com)11/24 Report--

On August 18, the China Computing Conference opened grandly in Yinchuan. At the "Intelligent Lenovo AI Computing Strategy and New AI Server launch Conference", Lenovo launched two new AI server products-- Lenovo WA7780 G3 AI large model training server and Lenovo WA5480 G3 AI training server.

These two servers are new AI-oriented products that can help build greener and more efficient artificial intelligence data centers to meet customers' computing needs ranging from training to reasoning.

AI intelligence, green empowerment: build an AI-oriented computing infrastructure and take the lead in China's intelligent reform

In recent years, under the background of "building digital China", the "foundation" of China's computing infrastructure has developed rapidly, and the intelligent transformation of the industry has been further deepened. At the same time, according to the IDC report, with the outbreak of generative artificial intelligence applications, the demand for intelligent computing in various industries exceeds the general computing power for the first time, and AI computing power has become the main direction of the development of computing power.

At the beginning of this year, Lenovo released Lenovo's first localized brand of Lenovo server, which takes "local innovation, agility and efficiency" as the brand connotation to better meet the personalized needs of local customers in China. Through the AI-oriented computing infrastructure, Lenovo accelerates the entry of artificial intelligence into a wide range of industries and covers multiple scenarios, and finally achieves the integration of universality and wisdom, leading the computing industry to a new future of universal, wisdom and green.

At the meeting, Chen Zhenkuan, vice president of Lenovo Group and general manager of the server division of Lenovo's infrastructure business group in China, said: "at a time when general computing power is developing towards AI computing power, Lenovo's WA7780 G3 AI large model training server and Lenovo's WA5480 G3 AI training server will meet the needs of different types of AI application scenarios."

Lenovo WA7780 G3 AI large model training server: strong performance, flexible and efficient

Lenovo WA7780 G3 is a server specially built for AI big model training. AI computing power has become a hard threshold for enterprises to train and develop and use artificial intelligence. Lenovo asks that the WA7780 G3 AI large model training server is to break the predicament of AI computing power supply, and to provide strong arithmetic support for training.

● surging computing power

Lenovo WA7780 G3 AI large model training server is based on the fourth generation Intel ®Xeon ®extensible processor. The NVIDIA HGX H800 system is equipped with 8 H800 NVLink ®GPU, HBM3 high-speed video memory up to 640GB, and interconnection bandwidth between GPU up to 400GB/s through NVLINK ®. The AI computing power is increased by 3.44 times to 32p FLOPS AI computing power. The NVIDIA H800 GPU matches the fourth generation Tensor Core with Transformer engine (FP8 accuracy). Compared with the previous generation products, it can improve the training speed of large model AI by up to 9 times and the reasoning speed of large model AI by up to 30 times.

● flexible architecture

Lenovo asks the WA7780 G3 AI large model training server, in order to meet the needs of high-speed data communication between GPU servers under the AI super-large model training scenario, it supports IB, RoCE and other external network connection schemes. It can support up to 8 RDMA high-speed network cards and provide 3.2Tb / s aggregate bandwidth. It fully meets the cross-node communication needs of super-large model training parallel computing. At the same time, GPU Direct Storage (GDS) technology is adopted in the design of local storage. Through GDS technology, GPU and NVMe achieve fast and low-delay interconnection, which can effectively improve the loading speed of training data and improve the performance of large model training.

Energy saving and high efficiency of ●

Lenovo WA7780 G3 AI large model training server adopts triple independent air duct design in the product design, which is systematically optimized according to the heat dissipation characteristics of different components, which effectively reduces the heat dissipation power caused by the fan. Compared with the same level of products, Lenovo WA7780 G3 AI large model training server reduces power consumption by about 10%. At the same time, in order to meet the lower PUE requirements of the intelligent computing center in the future, this product has reserved the liquid cooling board design, which will support liquid cooling in the future, which will greatly save the extra energy consumption caused by the heat dissipation of the system in the training process of the large AI model, and the PUE value is expected to be less than 1.1.

Lenovo WA5480 G3 AI training and push integrated server: diversified, stable and reliable

From machine vision, recommendation system, natural language processing, to the more complex AI model, different AI application scenarios have different requirements for AI servers. The Lenovo WA5480 G3 AI training and push integrated server, which was unveiled at this press conference, is a 4U rack-mounted AI server in the Lenovo brand, which supports multiple computing power and rich ecology, can provide extremely reliable computing power for the training and reasoning of the AI model, and accelerate the landing of artificial intelligence in thousands of industries, thus promoting the intelligent transformation process of various industries and even the whole society.

● multivariate computing power

Lenovo WA5480 G3 AI training and push all-in-one server uses two fourth-generation Intel ®strong ®scalable processors to support the latest PCIe 5.0. With the PCIe extension, up to 10 multi-type, multi-brand AI accelerator cards, including the latest NVIDIA H800 GPU, can be supported. It can be flexibly applied to AI general model training, large model reasoning, AI generation, cloud games, scientific computing and other application scenarios, providing multiple computing power for various scenarios of AI.

● flexible Topology

Lenovo WA5480 G3 AI training and pushing integrated server also fully embodies the flexible design concept in the hardware design of CPU-GPU interconnection. Lenovo WA5480 G3 AI training and push integrated server is based on different AI workloads, and can provide customers with a variety of CPU-GPU interconnection methods, including cut-through, Balance and Common, avoiding the potential performance bottleneck and system efficiency decline caused by the single CPU-GPU data communication mode and workload mismatch. Combined with different kinds and different number of accelerator card choices, it really realizes the perfect matching of all kinds of complex scenes in AI.

● is extremely reliable.

Lenovo WA5480 G3 AI training and pushing integrated server takes full account of design redundancy and provides a safe and stable operating environment for AI computing. Support power supply Numbn redundancy, backup each other between power supplies within the node, and connect two power lines externally to ensure that the server runs online 24 hours a day. Support CPU, GPU, memory and other key components of power consumption monitoring, power consumption capping control, dynamic protection of the server running in a reasonable range of power consumption, ensure stable performance, and fan Numb1 redundancy and other security design to ensure uninterrupted operation of high ring temperature.

These two new AI server products are important parts of the Lenovo Wendy brand family, inheriting the characteristics of high performance, high reliability, high expansion and low power consumption of Lenovo servers, and have the brand connotation of "local innovation, agility and efficiency". In the future, Lenovo WA7780 G3 AI large model training server and Lenovo WA5480 G3 AI training and push integrated server will be multi-directional empowered with Lenovo's other AI infrastructure product portfolio, helping AI to cover AI applications. At the same time, with Lenovo's leading liquid cooling technology, Lenovo's AI-oriented computing infrastructure will continue to help AI power green power, casting a solid "cornerstone" for Puhui Computing.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.