As AI workloads continue to scale, the infrastructure supporting them must be more robust than ever. This is where Dell PowerScale and Nvidia SuperPOD come into play, offering a powerful combination of storage and compute capabilities to accelerate AI-driven operations.

“It’s no surprise that Ethernet is predominant across all data centers today,” said Darren Miller (pictured), director of vertical industry solutions, unstructured data storage, at Dell Technologies Inc. “We believe as AI workflows increase and become the predominant workflow throughout the data centers, customers are going to need this high bandwidth. The DGX SuperPOD design with PowerScale was designed so that we would offer and bring to our customers the first Ethernet-based storage fabric for DGX SuperPOD.”

Ahead of the “Making AI Real With Data” event on October 15, Miller spoke with Rob Strechay, principal analyst at theCUBE Research, during a CUBE Conversation. They discussed the combination of Dell PowerScale and Nvidia SuperPOD as a scalable, high-performance solution that addresses the growing demands of AI workloads. (* Disclosure below.)

Scalable architecture with Dell PowerScale

Unlike previous generations that relied on technologies such as InfiniBand, the PowerScale-SuperPOD combination delivers the first Ethernet-based storage fabric for AI workloads. This breakthrough simplifies integration with existing infrastructures and provides the bandwidth necessary to feed data-hungry AI systems, making it an ideal choice for businesses looking to scale their AI operations quickly and efficiently, according to Miller.

“The PowerScale F710 platform is a single rack unit, dense platform that’s clustered together and scales incrementally as you add nodes and additional nodes to the cluster architecture,” he said. “This, in combination with PowerScale’s ability to handle high levels of concurrent connections and performance as we scale out the cluster, offers a very dense, high-performance system to run large-scale infrastructure such as the AI infrastructures we’re seeing today.”

Several features of PowerScale make it particularly well-suited for AI workloads, including its ability to manage high levels of concurrent connections and powerful data reduction capabilities. This combination of scalability, performance and efficiency allows businesses to maximize their ROI in AI infrastructure, according to Miller. As AI continues to grow in prominence, the scalability of solutions such as PowerScale will be key to supporting increasingly complex models and workloads.

“In addition to PowerScale’s other core fundamental capabilities, like data reduction, we offer two-to-one data reduction capabilities,” Miller said. “[With] multi-protocol capabilities and multi-tenancy, we have one of the most secure storage platforms in the business today and the industry, and we offer the latest in power and cooling technologies.”

One of the most significant challenges in AI infrastructures is ensuring that GPUs — often the most expensive component — remain fully utilized. GPUs are fast at processing data, but to unlock their full potential, they must be fed data at equally high speeds. This is where Dell PowerScale’s high-performance storage architecture comes into play, according to Miller.

“For example, data is staged and prepared for the GPUs to then consume and ingest that data for training and fine-tuning operations,” he said. “As that is taking place, those are seen as large amounts of concurrent connections from hundreds to thousands of GPUs in some of these larger infrastructures. The storage system must be able to handle that level of concurrency and provide service requests to those GPUs as they’re accessing data.”

Here’s the complete interview with Miller:

(* Disclosure: Dell Technologies Inc. sponsored this segment of theCUBE. Neither Dell Technologies nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE