NetApp has updated its E-Series line with two all-flash models (EF50 and EF80) aimed at artificial intelligence (AI) training, inference, and high performance computing (HPC) workloads.
With this launch, the performance of these E-Series arrays is said to be 2.5x faster. The E-Series has long been a high-speed option in the NetApp portfolio for applications that require dedicated bandwidth rather than the advanced storage capabilities of the Ontap-based FAS and AFF lines.
The new array is built to address the “data starvation” problem seen in GPU-intensive environments where storage I/O does not keep GPU utilization at optimal levels. According to NetApp, EF80 delivers read throughput of over 100GBps and write throughput of over 57GBps, targeting checkpoint writes during generative AI (GenAI) training, for example.
A 2.5x performance improvement compared to the previous generation is a significant improvement over existing EF models. In terms of density, NetApp packs 1.5 petabytes of storage into a 2U chassis. This is a move designed to curb the ever-increasing power and cooling footprint of modern data centers.
Target the neo-crowd
Sandeep Singh, senior vice president and general manager of storage, NetApp.
“This includes AI use cases including AI training, AI inference, HPC workloads, and transactional database workloads. This includes enterprise as well as neocloud, sovereign AI cloud, and AI-powered manufacturing use cases.”
Singh said the EF series is intended to serve as a high-speed “scratch space” in hierarchical architectures, often combined with parallel file systems such as Luster and BeeGFS. This allows the E-Series to act as a high-performance engine on the front end, with a larger, more persistent data store behind it.
E-series tradition
The E-Series occupies a unique space within NetApp. Its DNA comes not from the company’s famous WAFL-based filer, but from its 2011 acquisition of Engenio (from LSI). This gave NetApp much-needed block storage capabilities and enabled it to support workloads for which Ontap’s additional capabilities were actually a hindrance.
The E-series is a quiet workhorse, often working with products from IBM, Dell, and Teradata. While NetApp’s main AFF/FAS product line focuses on unified storage with massive data reduction, the E-Series remains the go-to for specialized high-duty cycle applications. The SANtricity operating system focuses on raw performance and simplicity.
Power the rack
As data centers shift their focus from pure capacity to power consumption, E-Series density becomes a key defense. Singh pointed out that storage consumption is a “small fraction” of what a rack of GPUs consumes. By feeding the GPU more efficiently and reducing idle time, the end result is a more streamlined, if still power-hungry, AI infrastructure.
“Putting this into the context of GPUs, these arrays ensure that the GPUs don’t sit idle or starve, and if you can optimize that part, you reduce wasted power and improve GPU utilization, which is also important to our customers,” said Singh.
This release puts NetApp in increased competition with the likes of Pure Storage and its FlashArray//XL, as well as Dell’s Project Lightning. While Pure focuses on a unified architecture, NetApp’s strategy is “natural” to use E-series for raw throughput and keep ONTAP for general-purpose enterprise data management.
