Covering Disruptive Technology Powering Business in The Digital Age

Home > Archives > News > DDN Announces A3I Solutions: Parallel Data Platforms for AI
DDN Announces A3I Solutions: Parallel Data Platforms for AI


DataDirect Networks (DDN®) announced a set of data platforms for artificial intelligence (AI) and deep learning (DL). Engineered for the AI data center, DDN’s A3I™ (Accelerated, Any-Scale AI) solutions are optimized to handle the spectrum of AI activities concurrently from data ingest and preparation to training, validation and inference. Rigorously tested and integrated around a comprehensive set of widely-used DL frameworks, network topologies and purpose-built GPU computing appliances, A3I solutions provide:

  • Demonstrated single NVIDIA GPU system throughput of more than 38GB/s;
  • Demonstrated single container throughput of more than 10GB/s;
  • Parallel, optimized data paths between AI workloads and storage;
  • Extremely high data Ingest and data transformation rates
  • Extensive interoperability and performance testing that has been completed with widely-used deep learning frameworks, notably TensorFlow, Horovod, Torch, PyTorch, NVIDIA® TensorRTTM, Caffe, Caffe2, CNTK, MXNET and Theano.

“With a track record of selling high-performance data storage for data-intensive computing, DDN is well positioned to provide storage systems that customers need to make AI and deep learning projects successful,” said Tim Stammers, senior analyst at 451 Research. “In general, AI holds great potential to solve problems across multiple industries and use cases. However, AI algorithms require fast access to data, which makes the choice of storage systems a critical foundational consideration.”

These solutions allow for scaling either with flash or with a hybrid approach, offering flexible scaling according to need with the performance of flash or the economics of hard disk drives (HDDs). DDN’s A3I solutions are factory pre-configured and optimized for AI, making them easy to deploy and reducing deployment time for an AI-ready data center. DDN is also developing A3I based solutions that include containerized deployment options, fitting into fully integrated and optimized GPU compute platforms such as NVIDIA® DGX-1™ and other NVIDIA GPU-accelerated compute architectures.

“Data is the new source code for enterprises, thanks to AI and deep learning, allowing organizations to transform products, services and business operations. Our customers look to NVIDIA to help them accelerate their adoption of AI through the power of GPU computing. Our collaboration with DDN to build integrated solutions will simplify the deployment of AI in the enterprise,” said Jim McHugh, VP and GM of Deep Learning Systems at NVIDIA.

A3I solutions are powered by DDN’s new scalable parallel file system appliances, the AI200™, AI400™ and AI7990™. The AI200 and AI400 are scale-out all-flash storage platforms. The AI7990 is a hybrid flash and hard drive storage platform that leverages parallel access to flash and deeply expandable HDD storage. All appliances support a scale-out model with solutions that start at 2U and a few TBs and can scale up to 10s of PBs. The AI7990 also allows for flexible scaling with deep capacity layers, for when customers require economic expansion to cater to growing AI datasets.

“The era of AI and deep learning is driving spectacular innovation across the industries that DDN serves,” said Alex Bouzari, founder and CEO of DDN. “Our goals are to help these organizations get more from their data with superior efficiency, to help drive fundamental change to markets and to improve lives.”

“DDN’s storage technologies are ideally suited for our deep-learning platform – but what really sets DDN apart is its expertise in deploying an integrated solution, which dramatically reduced our time to deployment,” said Jeff Hsu, intelligence and logistics manager, Standard Cognition. “DDN’s knowledge extends well beyond storage. Its engineering and support teams have been true partners in our mission to revolutionize retail checkout through AI.