IBM updated its Storage Scale System 6000 with new software, support for higher-density quad-level cell (QLC) flash, and a new all-flash expansion enclosure, tripling the system’s maximum capacity to 47-petabytes (PBs) per rack.

The system is designed to handle data-hungry AI applications and, when paired with the Nvidia BlueField data processing unit (DPU), can support high-performance workloads at scale.

“An AI factory without data is like an engine without fuel; it simply can’t run,” Sam Werner, GM of IBM Storage, wrote in a blog about the system upgrade.

Key enhancements with the latest 7.0.0 software release include the tripled maximum capacity to 47 PB per rack, enabled by support for QLC flash SSDs in 30 terabyte (TB), 60 TB, and 122 TB configurations; support for mixed nonvolatile memory express (NVMe) triple-level cell (TLC) plus QLC drives; a data-acceleration tier using NVMe over fabrics (NVMe-oF) delivering up to 340 Gb/s throughput and up to 28 million input/output operations per second (IOPS); and enhanced data-protection frameworks including broader erasure-code support.

The new all-flash expansion enclosure is optimized for AI training, inference, high-performance computing (HPC) and data-intensive workloads. It integrates with BlueField-3 DPUs, supports up to 400 Gb/s network throughput, and in a two-unit form factor offers up to 3 PB of raw flash capacity per enclosure.

IBM said the system supports multitenant isolation at cluster, fileset, or file-system levels, targeting service-provider and AI-factory use cases.

“IT infrastructure teams are faced with unprecedented data challenges to scale AI while continuing to support all existing workloads,” Werner explained. “The enhanced IBM Storage Scale System 6000 addresses these challenges and enables globally dispersed teams to collaborate on data regardless of protocol, location, or format.”

The latest software version and the all-flash expansion enclosure will be available early next month.

The update follows earlier improvements from IBM targeted at AI applications.

In an update this year, the company’s Storage Scale 6.0.0 adopted the storage scale system data acceleration tier (DAT), which uses high-speed NVMe-oF technology to provide ultra-fast, low-latency storage access for real-time AI inferencing.

In an interview with SDxCentral from earlier this year, the company predicted that the AI inference push will spill over into hybrid architectures, seeing data held on-premises and the use of inference models associated with trained models in the cloud.