TACC Builds 1 EB Archive for AI Research with Versity ScoutAM

The Texas Advanced Computing Center (TACC), based at The University of Texas at Austin, is a leader in advanced computational research. Its newest system, Horizon, delivers nearly 400 petaflops of peak performance. Recently named the U.S. National Science Foundation’s (NSF) Leadership-Class Computing Facility (LCCF), Horizon is one the largest academic supercomputer dedicated to open-scientific research worldwide.

With a focus on fields like biosciences, healthcare, computer vision, and AI, Horizon is a pivotal tool for groundbreaking research, providing 10x performance improvements and a 100x increase in AI capabilities over its predecessor, Frontera. To support this new era of large-scale, data-driven research, TACC needed an archive designed for AI-scale capacity, performance, and reliability.

To support Horizon’s unprecedented AI and simulation output, TACC deployed Ranch, an exabyte-scale archive powered by Versity’s ScoutAM. Running across 13 Dell PowerEdge R760 servers, ScoutAM orchestrates the entire storage environment:

• 16 PB Dell ECS object cache
• Two Spectra TFinity Plus tape libraries with 20 LTO-9 drives (LTO-10 ready)
• Five Dell ME5 arrays supporting active workloads

ScoutAM provides the intelligent data management layer that unifies flash, object storage, and tape into a seamless archive. Through policy-driven automation, it handles large-scale data ingest, placement, and recall, ensuring Horizon’s rapidly growing datasets remain both immediately accessible and durably preserved. Its modular, exascale-ready architecture allows metadata, performance tiers, and capacity tiers to scale independently. This gives TACC a future-proof archive that can grow for decades without workflow disruption, allowing the system to expand effortlessly as Horizon’s AI and simulation demands continue to accelerate.

A major advantage for TACC is ScoutAM’s extended caching architecture, designed for modern AI workflows. By coordinating a high-speed flash tier with a large object-storage cache, ScoutAM ensures that active or recently generated data is automatically staged for fast access, while long-term data is written directly to economical, high-capacity tape. This performance-first design keeps researchers productive while preserving long-term data cost-effectively on tape.

ScoutAM also brings built-in resilience through its open, format-transparent design. All metadata needed to rebuild the archive is stored directly with the data itself, eliminating proprietary dependencies and ensuring long-term recoverability. With features like asynchronous remote replication and a simple, elegant disaster-recovery model, ScoutAM gives TACC confidence that its exabyte-scale archive will remain stable, secure, and future-ready for decades.