High-Performance Software-Defined Storage for AI Workloads

IBM Storage Scale Erasure Code Edition (ECE) is a software-defined storage solution with a high-performance parallel file system (GPFS). It has been validated to work with Supermicro’s Petascale storage servers to offer exabyte-scale capacity, ultra-low latency, parallel I/O, and tiered architectures to efficiently support AI training and inference workloads. It can enable seamless data sharing across Supermicro Petascale nodes, providing a cost-optimized, scalable, and resilient platform for AI. With policy-based data management, it automatically tiers data across various storage mediums within Supermicro’s Petascale storage system to optimize performance and cost. This makes IBM Scale ECE an ideal solution for organizations seeking the elasticity and manageability of software-defined storage with the performance characteristics of enterprise-class file systems.

Maximize GPU Utilization and Eliminate Bottlenecks

IBM Storage Scale delivers parallel file system performance that minimizes idle GPU time caused by storage bottlenecks during AI training and inference workloads. When deployed on Supermicro Petascale servers, the solution provides the high-throughput, low-latency infrastructure needed to keep expensive GPUs running at peak efficiency. For AI and HPC teams, the platform's acceleration capabilities are particularly valuable in reducing wasted GPU hours caused by storage I/O waits.

Accelerate AI Workflows with Intelligent Tiering

IBM Storage Scale on Supermicro Petascale Servers support multi-flash tiering with NVMe TLC and QLC drives, allowing organizations to align cost, performance, and endurance with specific workload profiles. For AI inference, IBM Storage Scale enables KV cache sharing across thousands of GPU servers, delivering up to 50x+ speedup in time-to-first-token for large language models compared to recomputing.

Unify Data Across Distributed Environments

IBM Storage Scale provides a single global namespace that allows all GPU servers to access the same datasets simultaneously, eliminating data copying and version control issues across distributed AI teams. When deployed on Supermicro's Petascale storage servers, this disaggregated architecture allows you to independently scale compute and storage resources based on workload demands.
  • AI Data Ingest and AI Model Training
  • AI Model Tuning and AI Inferencing
  • High-Performance Computing (HPC)
  • Data-Intensive Research & Engineering
  • IBM Storage Scale Platform
    1. High-Performant File, Object and Vector Data Access
      (POSIX, GPUDirect, Container CSI, S3, SMB, NFS, HDFS)
    2. Secure Data Management and Transformation
      (Automated placement, movement, and vectorization)
    3. Seamless, Integrated Tiering
      (NVMe, SSD, NL-SAS, Tape)
      Seamless, Integrated Caching
      (Accelerate Remote File and Object Data)
Supermicro Petascale storage server for IBM Storage scale Code Edition
Supermicro Petascale Storage Server

Featured Products

Supermicro 1U Petascale E3.S Storage Server for IBM Storage Scale ECE solution

1U Petascale E3.S Storage Server

ASG-1115S-NE316R
Learn more
Resources