IBM Boosts Storage Scale System 6000 to 47PB, Targets AI Infrastructure Demands

Details

• IBM has upgraded its Storage Scale System 6000, tripling the maximum capacity to 47PB per rack and introducing Storage Scale 7.0.0 software, launching December 9, with new hardware following on December 12.
• The enhanced system is tailored for AI, HPC, and analytics workloads, supporting NVIDIA BlueField-3 DPUs and Spectrum-X Ethernet to accelerate model training and inference performance across edge, core, and cloud environments.
• Support for QLC NVMe SSDs in 30TB, 60TB, and 122TB variants, alongside existing TLC flash, allows for multi-flash tiering and a new data acceleration tier, achieving up to 340GB/s throughput and 28 million IOPS for demanding AI inference tasks.
• The new All-Flash Expansion Enclosure offers up to 100GB/s throughput and 3PB raw capacity per 2U unit, supporting up to 26 QLC drives and four BlueField-3 DPUs, enabling massive scalability within standard racks.
• Wider erasure coding options and multi-tenant configuration at various levels improve data protection, efficiency, and resource isolation, meeting the needs of large AI deployments and service providers.

Impact

With this update, IBM cements its leadership in high-density, high-performance storage for enterprise AI, aiming to tackle data movement bottlenecks in large-scale AI factories. The adoption of QLC flash signals a move toward affordable, expansive storage as a baseline in the AI arms race. Competing providers are likely to push similar innovations as demand for foundation model infrastructure accelerates.