BLOG

VelocityAI enables chief data officers and data scientists to accelerate AI and deep learning pipelines to drive faster ingest and quicker outcomes.

 

Data driven applications – Data has become the new source code

Digital transformation initiatives and data-driven applications are forcing many in automotive, healthcare, FSI and oil and gas to use AI and cogntive computing applications that leverage machine learning (ML) and deep learning (DL) data pipelines. IDC predicts that by 2021 75% of commercial enterprise applications will use AI, over 90% of consumers will interact with customer support bots, and over 50% of the new industrial robots will leverage AI.

 

The Challenge

The accuracy and predictability of these data pipelines is only as good as the data-sets they train on and effectiveness is only as good as non-blocking, continuous access to these data-sets. NVIDIA^®GPU supercomputers provide massively parallel compute layer, with 1 PFLOPs of computing power per DGX-1, however storage layer needs to match this massive parallelism to keep the GPU’s busy. Additionally different stages with data pipelines have distinct storage requirements of massive ingest bandwidth, mixed read/write handling for small and large files and ultra low latency.

Architectures using direct attached storage (DAS) will limit the performance and data mobility of the GPU, while existing all-flash storage lack the sustained performance to deliver timely insights at scale. This means business and IT leaders must reconsider how they architect their storage stacks and make purchasing decisions for these new workloads. VelocityAI delivers an industry best of breed solution for customers moving to AI and cogntive computing.

 

 

The solution – VelocityAI

VelocityAI marries the massive parallelism provided by  NVIDIA^®GPU supercomputers to the massive parallelism provided by Vexata Accelerated storage, removing I/O bottlenecks, poor utilization, and data silos frequently seen with these new AI/ML workloads. Line-of-business users as well as IT leaders can use AI/ML/DL workflows in their existing deployments that scale, accelerate and derive actionable business insights, thus democratizing cognitive computing.

VelocityAI solution comprises of NVIDIA^®DGX-1 GPU supercomputers, Mellanox 100 GbE switches and Vexata VX-100FS Accelerated File storage. The solution can scale from one DGX-1 server, all the way to four DGX-1 servers, while storage can scale from three blades all the way to 16 blades. Imagenet benchmarks ran on VelocityAI and demonstrated that VelocityAI is capable of handling all storage requirements of a pipeline – massive ingest, ETL, neural net modeling, training and inferencing stages. It provides highest bandwidth for each pipeline stage as well as the highest images/secs metrics.

VelocityAI Solution brief

VelocityAI Reference architecture

 

Customer case study

The use case was cancer anamoly detection using NVIDIA^®DGX-1 for advanced computer vision on hyperspectrometry images. The existing solution was based on Direct attached storage (DAS) architecture and took 20 days for CNN Neural net training. Using VelocityAI the customer was able to bring down the training time to 6 hours, showing an 80x improvement. Additionally VelocityAI was able to meet the distinct needs to hyperspectrometry image ingest, ETL processing into tensors, chosing the right neural net and then training phase.

 

Use cases

VelocityAI is ideal for deep learning, computer vision, and machine learning use cases, within the following verticals

• Healthcare
  • Medical Imaging
  • Pharma
  • Lifesciences/Genomics
  • Cancer detection

• Finance
  • Risk Analytics / Fraud Analytics / Anti-money laundering
  • Trading systems
  • Insurance

• Federal
  • High Performance Computing
  • Object Recognition
  • Voice/Video/Speech processing

• Oil and Gas
  • Simulations/Explorations

• Manufacturing/ Industry 4.0
  • Predictive Maintenance
  • Autonomous cars

• Retail
  • Price Optimization
  • Supply Chain

 

Routes to Market

VelocityAI will be sold through VAR’s with specialized AI/ML practices, who are part of Vexata partner program. GPL Technologies is one of the leading VAR’s in this area and is an elite NVIDIA^®partner.

 

Promotional activities

VelocityAI will be on display at AI SummitSan Francisco from September 19-20th, along with its channel partner GPL Technologies at booth #805. According to GPL technologies, VelocityAI provides them the flexibility, technology leadership and breakthrough economics to build different promotional bundles with Vexata accelerated storage.

Learn more at the webinar https://www.vexata.com/resource/turbocharge-data-pipelines-velocityai/