The new 400gbps Infiniband switch and network platform help AI, data analysis and high-performance computing (HPC) applications achieve the unity of security, cloud native, multi tenant and bare metal performance
Santa Clara, California, USA – GTC Conference – November 9, 2021 – NVIDIA today released NVIDIA quantum-2 – a new generation of Infiniband network platform, which provides cloud computing providers and supercomputing centers with extreme performance, extensive access capability and strong security.
NVIDIA quantum-2 platform is the 400gbps Infiniband network platform, including NVIDIA quantum-2 switch and connectx-7 ® Network card, bluefield-3 ® Data processor DPU (data processor) and all software supporting this new architecture. It is also by far the most advanced end-to-end network platform.
The launch of NVIDIA quantum-2 platform comes at a time when more and more supercomputing centers are opening up to the majority of users, including many external users. At the same time, global cloud service providers also begin to provide more supercomputing services for millions of their customers.
NVIDIA quantum-2 platform is tailored for the high demand applications of the above two trends. With its cloud native technology, this product provides high throughput of 400gb / s per second and advanced multi tenant support functions, which can meet the needs of many users.
Gilad shainer, senior vice president of NVIDIA network, “Today, the demands of Supercomputing Center and public cloud are moving towards integration – they must provide the highest possible performance for the new generation of high-tech computing (HPC), AI and data analysis applications. At the same time, they should safely isolate applications and respond to users’ different needs for traffic. With NVIDIA quantum-2 Infiniband platform, modern data center can turn this vision into reality.”
Performance and cloud native functions of NVIDIA quantum-2 platform
With a high throughput of 400 Gbps per second, NVIDIA quantum-2 Infiniband has doubled the network speed and tripled the number of network ports. While improving the performance by three times, it also reduces the number of switches required for the data center network by six times. At the same time, the energy consumption and space of the data center are reduced by 7%.
NVIDIA quantum-2 platform realizes the performance isolation between multiple tenants, which makes the behavior of one tenant not interfere with other tenants. At the same time, by using the advanced congestion control mechanism based on telemetry and supporting cloud native, it ensures reliable data throughput and is not affected by the peak demand of users or applications.
NVIDIA quantum-2 sharpv3 network computing technology can provide AI applications with 32 times more acceleration engines than the previous generation products, with the help of NVIDIA UFM ® Cyber AI platform will provide advanced Infiniband network management functions for the data center, including predictive maintenance.
NVIDIA quantum-2 platform integrates a nanosecond precision timing system, which can synchronize distributed applications, such as in database processing, and help reduce waiting and idle time. This new feature makes the cloud data center part of the telecommunications network and can host 5g wireless services defined by software.
Quantum-2 Infiniband switch
The core of quantum-2 platform is the brand-new quantum-2 Infiniband switch. Its main chip adopts 7-nm process and contains 57 billion transistors, slightly more than NVIDIA A100 GPU with 54 billion transistors.
It has 64 400gbps ports or 128 200gbps ports, and will provide switch systems with different port numbers, up to 2048 400gbps ports or 4096 200gbps Ports – about five times higher than the previous generation quantum-1 in terms of switching capacity.
With the advantages of network speed, switching capacity and high scalability, it has become an ideal choice for building the next generation of giant high-performance computing system.
Many leading infrastructure and system manufacturers in the world now support NVIDIA quantum-2 switches, including ATOS, DataDirect networks (DDN), Dell, excelro, gigabyte, HP, IBM, Inspur, Lenovo, Penguin computing, QCT, ultramicro, vast data and wekaio.
Quantum-2 connectx-7 and bluefield-3
NVIDIA quantum-2 platform provides two network options on the host side, NVIDIA connectx-7 network card and NVIDIA bluefield-3 DPU Infiniband.
Connectx-7 is designed based on 7-nanometer process and contains 8 billion transistors. Its data transmission rate is twice that of NVIDIA connectx-6, the world’s leading high-performance computing network chip. It also enables RDMA and gpudirect ® Double the performance of storage, gpudirect RDMA and network computing. The connectx-7 prototype will be available in January next year.
Bluefield-3 Infiniband is also designed with 7-nanometer process, including 22 billion transistors and providing 16 64 bit arm CPUs to unload and isolate various data center infrastructure services. The bluefield-3 prototype will be available in May next year.