New NVIDIA DGX-2 artificial intelligence system includes Mellanox InfiniBand and Ethernet offerings

Mellanox Technologies, supplier of end-to-end smart interconnect solutions for data center servers and storage systems, announced Monday that its InfiniBand and Ethernet solutions have been chosen to accelerate the new NVIDIA DGX-2 artificial intelligence (AI) system.

DGX-2 is initial 2 Petaflop system that combines sixteen GPUs and eight Mellanox ConnectX adapters, supporting both EDR InfiniBand and 100 gigabit Ethernet connectivity. The technological advantages of the Mellanox adapters with smart acceleration engines enable the highest performance for AI and deep learning applications. The embedded Mellanox network adapters provide overall 1600 gigabit per second bi-directional data throughout, which enables scaling up AI capabilities for building large deep learning compute systems.

The DGX-2 platform, together with the DGX-1 platform that is also based on Mellanox network adapters, provide a set of options for AI companies, users and developers to build the next generation of AI products and services.

DGX-2 delivers a ready-to-go solution that offers the fastest path to scaling up AI. DGX-2 simplifies speed of scaling-up AI with flexible networking options for building the largest deep learning compute clusters, combined with virtualization features that enable improved user and workload isolation in shared infrastructure environments. With an accelerated deployment model and an architecture purpose-built for ease of scale, your team can spend more time driving insights and less time building infrastructure.

With DGX-2, model complexity and size are no longer constrained by the limits of traditional architectures. Users can now adopt model-parallel training with the NVIDIA NVSwitch networking fabric, which is included in the new platform to offer 2-petaFLOPS GPU accelerator with 2.4 TB/s of bisection bandwidth, delivering a 24 times increase over prior generations.

“The DGX-2 is the first single server capable of delivering two petaflops of computational power that combines 16 fully interconnected GPUs for 10X the deep learning performance,” said Jim McHugh, VP and GM of Deep Learning Systems at NVIDIA. “Mellanox InfiniBand and Ethernet solutions enable us to give maximum flexibility and performance to customers who build out large-scale clusters of DGX-2 systems.”

“We are excited to collaborate with NVIDIA and to bring the performance advantages of our EDR InfiniBand and 100 gigabit Ethernet to the new DGX-2 Artificial Intelligence platform,” said Gilad Shainer, vice president of marketing at Mellanox Technologies. “Doubling the network throughput as compared to previous systems to provide overall bi-directional 1600 gigabit per second data speed enables the DGX-2 platform to analyze growing amounts of data, and to dramatically improve Deep Learning application performance.”


Leave a Reply

WWPI – Covering the best in IT since 1980