Overview

AI workloads create the need for simple, fast, and affordable storage that can feed high throughput GPUs with the entirety of large datasets. Meet VAST at NVIDIA’s GTC’21 to learn how Universal Storage helps customers all over the world simplify their AI infrastructure and transform their data science initiatives with petabyte-scale, state of the art flash storage at 1/10th the cost.

Sessions

SS33082: Demystifying Storage Access Patterns to Unlock Efficient and Scalable AI

session-image
session-logo

Sven Breuner

,Field CTO, VAST Data

The rise in adoption of AI is spurring an infrastructure revolution in data centers all around the globe. While GPUs are at the center of any AI conversation, understanding key storage system characteristics is equally paramount to successfully and efficiently scale AI initiatives. In this session, we will characterize what makes Deep Learning I/O unique and why traditional storage systems fail to keep up with these modern AI workloads. We will also discuss strategies for overcoming storage bottlenecks and explore new technologies such as NVIDIA GPUDirect Storage. Lastly, we will present "elbencho," a new open-source storage benchmark that can help data scientists validate the impact of various data formats and access patterns on storage systems without needing to modify their AI applications.

View Session

SS33083: Leveraging NVMe Fabrics and Bluefield to Build Hyperscale Flash Archives

session-image
session-logo

Jeff Denworth

,Co-Founder and CMO, VAST Data

The idea of the all-flash data center has long been an unobtainable dream for data-driven organizations. That said, the future is bright. Advances in modern networking and DPUs have made it possible to develop a next-generation data center storage architecture that borrows from web scale disaggregation principles to exploit never-before-seen efficiencies in flash storage. In this session, VAST Data will highlight how NVMe-over-Fabrics lays the foundation for a new architecture that is capable of global QLC flash management, global compression and low-overhead global erasure codes - all contributing to a new hyperscale flash economic model that makes it possible for flash to be affordable for all data.

In this same session, VAST Data will preview some of its work with NVIDIA’s Bluefield SmartNICs and Data Processing Units (DPUs) in building more efficient, modular and highspeed NVMe over Fabrics targets.

View Session

SS33080: Smart Coupling of Next Generation A.I. Computing with Hyperscale Low-Cost Flash

session-image
session-logo

Jeff Denworth

Co-founder and CMO, VAST Data

session-image
session-logo

Matt Hull

Vice President, Global AI Data Center Solutions Sales, NVIDIA

session-image
session-logo

Eric Burgener

Vice President, Research, IDC

Artificial intelligence (AI) is becoming mainstream across a wide range of industries. From self-driving vehicles, to genomics research, to financial data analysis, AI is rapidly weaving through our world in ways both seen and unseen.

The future success of AI initiatives will be dictated by IT organizations’ ability to quickly deliver simple-to-use yet highly-optimized infrastructure to large teams of data scientists. Productive planning for new AI projects requires IT professionals take a fresh approach in building and deploying emerging compute and storage platforms.

Hosted by IDC, in this session, you’ll hear industry experts from NVIDIA and VAST Data discuss:

  • Emerging AI trends’ impact on the future of IT infrastructure
  • AI compute and storage requirements and modern infrastructure approaches
  • How NVIDIA DGX systems and VAST Data’s Universal Storage bring together performance and simplicity fit for any AI data center project
View Session

S32134: Coupling Next Generation AI with Hyperscale Low-Cost Flash using High Performance Ethernet Networking

session-image
session-logo

Matt Hull

Vice President, Global AI Data Center Solutions Sales, NVIDIA

session-image
session-logo

Alon Horev

Vice President, Technology, VAST Data

session-image
session-logo

Rob Davis

Vice President, Storage Technology, NVIDIA

In this talk, you’ll hear industry experts from NVIDIA and VAST Data discuss:

  • Emerging AI trends and the impact on future IT infrastructure.
  • AI compute, storage, and network requirements and modern infrastructure approaches.
  • How NVIDIA Networking connects VAST Data Universal Storage with DGX Pod for performance and simplicity to fit any AI data center project

SS33155: Accelerating Modeling Velocity

session-image
session-logo

Ian Ferreira

Chief Product Officer, AI, Core Scientific

session-image
session-logo

John Mao

Head of Business Development, VAST Data

Data scientists are highly skilled individuals charged with, and trained in, the art and science of uncovering patterns from massive amounts of data. They should be spending their time building strategic models that offer deep insights into today’s biggest challenges instead of selecting and building infrastructure to support their needs. This session looks at the best “MLOps” tools and processes for accelerating AI modeling velocity so data scientists can maximize personal and organizational efficiency.

View Session
Virtual Conference

VAST Data at GTC21