Dell bets liquid-cooled, software-defined storage can keep AI factories fed

Dell bets liquid-cooled, software-defined storage can keep AI factories fed
AI factories are becoming so powerful and crowded that storage is now just as important as computing power in deciding how quickly they can turn data into tokens. For Dell Technologies Inc., the solution starts with turning its PowerEdge servers into flexible storage systems controlled by software. These systems can change their roles as workloads change.
As companies move from training AI models to using them for predictions and AI agents, they need results faster. Dell has spent the year expanding its AI Data Platform with a semantic layer designed for AI agents. Its AI factory plans now cover much more than computing and graphics processing units (GPUs), according to George Papadopoulos, senior vice president of product and engineering at Dell.
“We look at our AI factory not just as a set of CPUs and GPUs, but networking becoming more and more important, particularly around speed, reliability of storage, and the ability to take your raw data and generate context and knowledge from it,” Papadopoulos said.
Papadopoulos spoke with theCUBE Research’s John Furrier and Rebecca Knight at SC26. The interview took place during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. (* See the disclosure below.)
PowerEdge storage designed to adapt to changing AI workloads
Customers want to buy storage for their needs today and be able to use it for something else later. For example, a customer might start with file storage and later switch to object storage or parallel file storage as their workloads change. AI agent workloads rely more on small input and output operations than on streaming large amounts of data. This means the best type of storage can change quickly, Papadopoulos explained. Dell’s solution is Dell Exascale Storage, software that customers license and run on Dell servers.
“It’s our really purpose-built software-defined solution for allowing us to take a PowerEdge-based node and … with flexible licensing, allow you to deploy a file persona, a block persona, an object persona, or a parallel file persona, and allow that flexibility to change over time as your needs change,” Papadopoulos said.
The growing number of powerful systems is also changing how data centers need to be cooled. Today's AI systems use so much power that customers are planning data centers that are 100% liquid-cooled, Papadopoulos said. Storage systems must now meet this same cooling requirement.
“So the conversations have really gotten down to how can we most effectively provide a direct liquid cooled solution and the storage that means even liquid cooling the drives, which is something that hasn’t been done as of yet,” Papadopoulos said.
At SC26, Dell announced a storage system that is 2U high and holds 32 drives. It uses liquid cooling for every part of the system. Dell calls it the first direct liquid-cooled storage system designed for rack-scale AI. The system runs as an Exascale system and can use any storage type, or persona. Keeping the software separate from the hardware also helps Dell bring new generations of PowerEdge servers to market faster. This flexibility is important for neocloud operators, who need a way to move away from each purchase if their needs change, Papadopoulos noted.
“That allows them to make an upfront purchase and know that they won’t be locked into one particular storage type,” Papadopoulos said, “or have a set of data silos that they can’t leverage as they go forward.”
Here is the full video interview, part of SiliconANGLE’s and theCUBE’s coverage of SC26:
(* Disclosure: Dell Technologies Inc. paid to sponsor this part of theCUBE. Dell Technologies Inc. and other sponsors do not control the editorial content on theCUBE or SiliconANGLE.)
