Skip navigation
IBM Elastic Storage System 3500 IBM

IBM Launches Storage Server for AI Workloads

ESS 3500 is up to 12% faster than previous generation systems.

IBM has unveiled the Elastic Storage System 3500 – a 2U storage server designed specifically for AI training workloads.

The new unit houses 24 drive bays and has a maximum raw capacity of 368TB of NVMe storage.

Making use of Spectrum Scale, IBM’s high-performance clustered file system software, the ESS 3500 can achieve up to 91GB/s of throughput.

The system is designed to "break performance barriers" for AI – with increased throughput enabling GPUs to solve AI problems faster.

The new offering can be combined with other IBM storage units – including previous entries in the ESS line. It is also compatible with non-IBM storage. Given its improvements to GPU output, it could be paired with Nvidia DGX A100 systems – which are GPU-focused but lack storage options.

The solution comes with Spectrum Scale installed on a pre-configured system, with updates delivered at speed through containerized software.

ESS 3500 key specs

System features

  • Dual 1-socket Storage Controllers, Active/Active
  • 1024 GB memory
  • De-Clustered RAID supporting erasure coding schemas: 3-way replication, 4-way replication, 4+2P, 4+3P, 8+2P, 8+3P


  • AMD 7642 48 core processor
  • Sequential read performance up to 91GB/S


  • Four x16 PCIe Gen4 adapter slots

Drive support

  • 12 or 24 NVMe SSDs (3.84TB, 7.68 TB or 15.36 TB)


  • ESS software
  • IBM Spectrum Scale for ESS
  • Red Hat Enterprise Linux (RHEL) 8.4

Full specifications can be found here.

This story originally appeared on AI Business, a Data Center Knowledge sister publication. To stay up-to-date with artificial intelligence news, subscribe to the AI Business newsletter.

TAGS: Storage IBM
Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.