Sign up for our newsletter and get the latest HPC news and analysis.
Send me information from insideHPC:


DDN feeds NVIDIA DGX Servers 33GB/s for Machine Learning

Today DDN announced that its EXAScaler DGX solution accelerated client has been fully integrated with the NVIDIA DGX Architecture. Using the EXAScaler ES14KX high-performance all-flash array, the new solution smashed existing records by demonstrating a massive 33GB/s of throughput to a single NVIDIA DGX-1.

DDN is focused on solutions that enable customers to radically accelerate their AI and Machine Learning initiatives to gain competitive advantage sooner and to scale flexibly,” said James Coomer, vice president of product management at DDN. “We have taken a holistic approach to this solution, ensuring that the DGX platform is maximally utilized and that customers can extract deeper insights more quickly.”

By supplying this groundbreaking level of performance, DDN enables customers to greatly accelerate their Machine Learning initiatives, reducing load wait times of large datasets to mere seconds for faster training turnaround. The DDN EXAScaler family of products, along with the EXAScaler DGX Solution provides a purpose-built platform uniquely tuned to the needs and characteristics of data flows found in AI and Machine Learning workloads. EXAScaler further improves the flexibility of the AI datacenter by allowing massive ingest rates and cost-effective capacity scaling.

At GTC18, DDN has demonstrated an EXAScaler DGX Solution based on the ES14KX all-flash array that achieves more than 33GB/s throughput to a DGX-1 server and more than 250,000 random read 4K IOPS, all without stressing the host CPU on the DGX server. Exceeding 10GB/s and 100,000 IOPS to a single container gives more than 12X the performance attainable from competing NFS solutions and enables data scientists to fully extract the performance from the flash layer rather than being limited by wire protocols.

DDN’s EXAScaler solutions enable organizations to overcome the challenges and inefficiencies caused by I/O bottlenecks and allow customers to realize true flash-cache economics by providing a storage architecture that is ideally suited to Deep Learning and Machine Learning requirements. DDN recently announced Standard Cognition is leveraging EXAScaler to revolutionize the consumer shopping experience with fully autonomous check-out.

Sign up for our insideHPC Newsletter

Leave a Comment

*

Resource Links: