DDN Storage: 33GB/s Performance to NVIDIA DGX Servers to Accelerate Machine Learning and AI Initiatives

DataDirect Networks, Inc. (DDN) announced its EXAScaler DGX solution, a solution that delivers performance using a optimized, accelerated client integrating tightly and with the Nvidia Corporation‘s DGX architecture.

Using the EXAScaler ES14KX performance all-flash array, the solution smashed existing records by demonstrating 33GB/s of throughput to a single Nvidia DGX-1.

Nvidia DGX-1

By supplying this level of performance, the company enables customers to accelerate their machine learning initiatives, reducing load wait times of large data sets to meres for faster training turnaround. The EXAScaler family of products, along with the EXAScaler DGX solution, provides a purpose-built platform 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.

“DDN is focused on solutions that enable customers to accelerate their AI and Machine Learning initiatives to gain competitive advantage sooner and to scale flexibly,” said James Coomer, VP, product management, 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.“

At GTC18, the company 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 IO/s, all without stressing the host CPU on the DGX server. Exceeding 10GB/s and 100,000 IO/s to a single container gives more than 12X the performance attainable from competing NFS solutions and enables data scientists to extract the performance from the flash layer rather than being limited by wire protocols.

EXAScaler solutions enable organizations to overcome the challenges and inefficiencies caused by I/O bottlenecks and allow customers to realize flash-cache economics by providing a storage architecture suited to deep learning and machine learning requirements. The company recently announced Standard Cognition is leveraging EXAScaler to revolutionize the consumer shopping experience with autonomous check-out.

The firm showcased its EXAScaler storage solutions this week at the GPU Technology Conference in San Jose, CA.