The Challenge of Scaling to Meet the Demands of Modern AI and Deep Learning

Deep neural networks are rapidly growing in size and complexity, in response to themost pressing challenges in business and research. The computational capacityneeded to support today’s modern AI workloads has outpaced traditional data centerarchitectures. Modern techniques that exploit increasing use of model parallelismare colliding with the limits of inter-GPU bandwidth, as developers build increasinglylarge accelerated computing clusters, pushing the limits of data center scale. A newapproach is needed - one that delivers almost limitless AI computing scale in order tobreak through the barriers to achieving faster insights that can transform the world.

Performance to Train the Previously Impossible

Increasingly complex AI demands unprecedented levels of compute. NVIDIA®DGX-2™ is the world’s first 2 petaFLOPS system, packing the power of 16 of theworld’s most advanced GPUs, accelerating the newest Deep Learning model typesthat were previously untrainable. With groundbreaking GPU scale, you can trainmodels 4X bigger on a single node. In comparison with legacy x86 architectures,DGX-2’s ability to train ResNet-50 would require the equivalent of 300 servers withdual Intel Xeon Gold CPUs costing over $2.7 million dollars.

NVIDIA NVSwitch—A Revolutionary AI Network Fabric

Leading edge research demands the freedom to leverage model parallelism andrequires never-before-seen levels of inter-GPU bandwidth. NVIDIA has createdNVSwitch to address this need. Like the evolution from dial-up to ultra-high speedbroadband, NVSwitch delivers a networking fabric for the future, today. With NVIDIADGX-2, model complexity and size are no longer constrained by the limits of traditionalarchitectures. Embrace model-parallel training with a networking fabric in DGX-2 thatdelivers 2.4TB/s of bisection bandwidth for a 24X increase over prior generations. Thisnew interconnect “superhighway” enables limitless possibilities for model types thatcan reap the power of distributed training across 16 GPUs at once.

AI Scale on a Whole New Level

Modern enterprises need to rapidly deploy AI power inresponse to business imperatives, and need to scaleoutAI, without scaling-up cost or complexity. We’vebuilt DGX-2 and powered it with DGX software thatenables accelerated deployment, simplified operations– at scale. DGX-2 delivers a ready-to-go solution thatoffers the fastest path to scaling-up AI, along withvirtualization support, to enable you to build your ownprivate enterprise grade AI cloud. Now businessescan harness unrestricted AI power in a solution thatscales effortlessly with a fraction of the networkinginfrastructure needed to bind accelerated computingresources together. With an accelerated deploymentmodel, and an architecture purpose-built for ease ofscale, your team can spend more time driving insightsand less time building infrastructure.

Enterprise Grade AI Infrastructure

If your AI platform is critical to your business, youneed one designed with reliability, availability andserviceability (RAS) in mind. DGX-2 is enterprise-grade,built for rigorous round-the-clock AI operations,and is purpose-built for RAS to reduce unplanneddowntime, streamline serviceability, and maintainoperation continuity.Spend less time tuning and optimizing and moretime focused on discovery. NVIDIA’s enterprise-gradesupport saves you from the time-consuming job oftroubleshooting hardware and open source software.With every DGX system, get started fast, train faster, andremain faster with an integrated solution that includessoftware, tools and NVIDIA expertise.
 

Powered by NVIDIA TESLA V100