NVSwitch on the DGX-2 is a super crossbar switch which greatly increases the application performance in several ways. First, it increases the problem size capacity traditionally limited by a single GPU's memory to the aggregate DGX-2 GPU memory of 512 GB. Second, NUMA-effects of traditional multi-GPU servers are greatly reduced, growing memory bandwidth with the number of GPUs. Finally, ease-of-use is simplified as apps written for a smaller number of GPUs can now be more easily ported with the large memory space.
Get an inside look at the world's most powerful AI system, NVIDIA DGX-2. Explore the design and hardware architecture that enables sixteen Tesla Volta GPUs to operate as one giant GPU. Find out how NVIDIA DGX-2 can enable you to explore and solve the most complex AI challenges.
This talk will present the results of running the following Graph500 and DARPA Graph Challenge benchmarks and highlight the improvements over other platforms: BFS Graph500
• Single Source Shortest Paths Graph500
• PageRank Pipeline Graph Challenge
• Triangle Counting Graph Challenge
• K-Truss Graph Challenge The tremendous performance advantages of the DGX-2 platform for deep-learning has recently gained a lot of publicity. However, that is not the only analytic environment that can take advantage of the DGX-2 architecture. Having sixteen fully connected 32GB Volta GPUs presents an intriguing platform for Graph Analytics. The 512GB of combined GPU memory and full NVLink connection between the GPUs offers a number of advantages over a distributed MPI-based approach.
This technical session will explore the objectives for building the DGX-2, along with the inspired, innovative technology and architecture used to eliminate traditional bottlenecks and to enable multi-GPU training at unprecedented scale. This talk led by the DGX product team will present on the following topics:
-the innovation and architecture found in NVSwitch, which enables the AI network fabric for the DGX-2 platform
-the design challenges and hardware architecture employed to enable 16 V100's to operate as one
-feature by feature walkthrough highlighting the most important innovations that accelerate deep learning workflow and training performance
-use cases that were previously unaddressable on a GPU platform, now solved with DGX-2
PSC's "Bridges" was the first system to successfully converge HPC, AI, and Big Data. Designed for the U.S. national research community and supported by NSF, it now serves approximately 1600 projects and 7500 users at over 350 institutions. Bridges emphasizes "nontraditional" uses that span the life, physical, and social sciences, engineering, and business, many of which are based on AI or AI-enabled simulation. We describe the characteristics of Bridges that have made it a success, and we highlight several inspirational results and how they benefited from the system architecture. We then introduce "Bridges AI", a powerful new addition for balanced AI capability and capacity that includes NVIDIA's DGX-2 and HPE NVLink-connected 8-way Volta servers.
This session presents an overview of the hardware and software architecture of the DGX-2 platform. This talk will discuss the NVSwitch hardware that enables all 16 GPUs on the DGX-2 to achieve 24x the bandwidth of two DGX-1V systems. CUDA developers will learn ways to utilize the full GPU connectivity to quickly build complex applications and utilize the high bandwidth NVLINK connections to scale up performance.
NVIDIA's DGX-2 system offers a unique architecture which connects 16 GPUs together via the high-speed NVLink interface, along with NVSwitch which enables unprecedented bandwidth between processors. This talk will take an in depth look at the properties of this system along with programming techniques to take maximum advantage of the system architecture.