Inspur has released five new AI servers that fully support the new NVIDIA® Ampere architecture. The new server can support 8 to 16 NVIDIA A100 Tensor Core GPUs, which can provide up to 4,000 teraflops of AI computing performance and up to 600GB/S of P2P bandwidth between GPUs.
San Jose, California, – On May 14, Inspur, the world’s leading AI computing infrastructure manufacturer, announced a comprehensive upgrade of its AI servers to fully support the new NVIDIA A100 Tensor Core GPU. .
The five AI servers released by Inspur will deal with various artificial intelligence computing scenarios, and can support 8 to 16 latest NVIDIA A100 Tensor Core GPUs. The third-generation Tensor Core cores in the A100 GPUs are faster, more flexible, and easier to use, enabling these servers to perform up to 4,000 teraflops, while enabling inter-GPU performance through next-generation NVIDIA NVSwitch technology. The P2P communication performance is doubled and up to 600GB/S. Stronger computing performance and higher interconnection bandwidth between GPUs can bring higher AI computing efficiency and support users to conduct AI training of larger data scale and more complex models. In addition, the NVIDIA A100 GPU has 40GB of HBM2 memory, and the memory bandwidth is increased by 70% to 1.6TB/sec, which can carry larger deep learning Model training.
The five AI servers released by Inspur will provide hardware design for a variety of application scenarios, including:
1. NF5488M5-D can accommodate 8 A100 GPUs in 4U space, and supports two current most mainstream general-purpose CPUs, which can provide users with leading AI computing technology and mature ecological support.
2. NF5488A5 supports 8 third-generation NVlink interconnected A100 GPUs and 2 PCI-E Gen4 CPUs, providing users with high-bandwidth data communication.
3. The NF5488M6 will provide an extremely optimized design system for large-scale HPC and AI computing users, and will integrate 8 A100 GPUs and 2 next-generation mainstream general-purpose CPUs that support NVSwitch high-speed interconnection.
4. NF5688M6 will provide the ultimate expansion performance for AI users, support 8 A100 GPUs and accommodate more than 10 PCIe Gen4 devices at the same time, and achieve a 1:1:1 balanced ratio of GPU computing, NVMe storage and NVIDIA Mellanox InfiniBand network .
5. NF5888M6 can realize high-speed interconnection of 16 A100 GPUs in a single machine through NVSwitch, and the single-machine performance can reach 4000 trillion times per second with INT8 precision, providing the most powerful AI super server for customers who pursue extreme performance.
Inspur also announced that its leading AI computing resource platform AIStation and automatic machine learning platform AutoML Suite will also simultaneously support the latest A100 GPU. AutoML provides flexible AI computing system resource scheduling and powerful AI model algorithm development support.
In addition, Inspur plans to add the new EGX A100 to its edge server product portfolio, offering enhanced security and unprecedented performance at the edge. The EGX A100 Convergence Accelerator combines Mellanox SmartNICs and GPUs powered by the new NVIDIA Ampere architecture so businesses can run AI at the edge more securely.
Liu Jun, general manager of AI & HPC of Inspur Group, said: “Inspur launched a full line of AI server products based on the NVIDIA Ampere architecture at the first time, providing global AI customers with a new generation of AI computing platforms specially designed for the ultimate innovation and optimization of application scenarios, further enhancing Inspur The most abundant AI server product array in the industry can better cope with the huge computing challenges brought by the current massive data explosion and complex models. This is also the industry’s most powerful AI computing power unit, which will effectively promote the intelligent computing center AI in construction and industry will be implemented.”
Paresh Kharya, Director of Accelerated Computing Product Management at NVIDIA, said, “The NVIDIA A100 Tensor Core GPU can provide users with unparalleled acceleration in various fields such as AI, data analysis and HPC. The Inspur AI server powered by the NVIDIA A100 GPU will help global users eliminate computing bottleneck, significantly reducing costs, energy consumption and space requirements for data centers.” Inspur is the world’s leading AI server manufacturer, with a rich array of AI products, and works closely with AI customers to help in voice, semantics, image, video, Search and other aspects have achieved an order of magnitude improvement in AI application performance. According to the IDC data report, Inspur AI servers have maintained a market share of more than 50% in China’s AI infrastructure infrastructure field for three consecutive years.