NetApp E-Series storage certified for Nvidia SuperPod

NetApp has organized its E-Series hardware with BeeGFS software to be certified with Nvidia’s SuperPod AI powerhouse.

The all-flash EF600/BeeGFS combo has been certified for use with the DGX SuperPod, a rack-based AI supercomputer with 20 to 140 DGX A100 GPU systems and HDR (200Gbit/s) InfiniBand network links. BeeGFS is an open-source, scale-out, clustered, parallel file system supported by ThinkParQ. The EF600 runs the SANtricity OS and delivers up to 44GB/sec throughput and up to 2 million random read IOPS with less than 250μs latency. 

Phil Brotherton, Netapp VP of Solutions and Alliances, said: “The NetApp and NVIDIA alliance has delivered industry-leading innovation for years, and this new qualification for NVIDIA DGX SuperPOD builds on that momentum.”

NetApp EF600
NetApp EF600

He added: “As performance and data demands are exploding, NVIDIA DGX SuperPOD, coupled with ONTAP AI platform and DGX Foundry AI service, ensures our customers get the best-in-class model training infrastructure, with the full support of AI industry leaders.” 

Nvidia’s DGX Foundry AI service is a cloud-hosted development system based on the DGX SuperPod and a dedicated high-availability (HA) pair of NetApp AFF A800 storage arrays – not the EF600/BeeGFS system. It is accessed via subscription and includes Nvidia’s Base Command software.

The DGX Foundry AI service does not support the EF600, and NetApp said: “With today’s announcement, organizations can also build their AI Centers of Excellence leveraging the power of DGX SuperPOD on premises in combination with NetApp [EF600] storage.” 

The EF600, first announced in August 2019, is a hot and dense box. It supplies up to 150,000 random read IOPS at less than 100μs latency and its 2RU chassis contains 24 x NVMe SSDs, 367TB of raw capacity. The system supports 100Gb NVMe over InfiniBand, 100Gb NVMe over RoCE, 200Gb NVMe over InfiniBand, 200Gb NVMe over RoCE, 32Gbit/s NVMe over FC, and 32Gbit/s FC.

NetApp’s AFF A800 is slower than the EF600, providing 1.3 million IOPS at 500μs latency and throughput of up to 34GB/sec with a high-availability pair.

NetApp and Nvidia are working with Wichita State University to provide a SuperPOD and HPC/AI Center of Excellence for applied learning students to experience supercomputing and high-performance computing environments.