Argonne's Aurora Storage Hits 30 Terabytes Per Second

Argonne Leadership Computing Facility published DAOS performance results from its 800-server storage configuration on the Aurora supercomputer.

Updated on Nov. 9, 2026 in Data Centers

Isometric editorial illustration of rows of tall matte-gray server cabinets in a clean data hall with small warm glow points. — Argonne National Laboratory's Aurora supercomputer storage system, built from 800 servers, reached data transfer rates of up to 30 terabytes per second in recent testing. AI Illustration. Upload story photo >

The Argonne Leadership Computing Facility released performance results for the DAOS storage system on its Aurora supercomputer, which runs an 800-server configuration. DAOS delivered data transfer rates of up to 30 terabytes per second during testing.

Why it matters

The results show how large-scale storage can keep pace with the compute demands of modern scientific simulation and AI workloads. For Illinois, home to the facility, the benchmarks document the capabilities of a major research computing installation.

DAOS reached a peak data transfer rate of 30 terabytes per second across 800 storage servers on the 10,624-node Aurora system. In the 2025 MLPerf Storage benchmark, a 128-server subset reached nearly 1 terabyte per second of writes and 600 gigabytes per second of reads.

The players

Argonne Leadership Computing Facility

The Argonne Leadership Computing Facility is a Department of Energy user facility in Illinois that operates the Aurora supercomputer for open science research.

The details

Argonne researchers ran two large HACC cosmological simulations on Aurora that wrote more than 85 petabytes of data through DAOS, retaining about 8 petabytes for scientific analysis. DAOS distributes data and metadata across storage servers without relying on a centralized metadata server.

January 2025: Aurora supercomputer went live.
2025: MLPerf Storage benchmark run on a 128-server subset of Aurora's DAOS system.
November 9, 2026: Argonne released the DAOS performance numbers.

The Tech Race

The DAOS results document how Aurora's storage scaled from 128 to 800 servers while retaining first place on the ISC26 IO500 Production list. The figures position the system among the highest-throughput installations for data-intensive scientific computing.

Researchers using Aurora can run simulations that write tens of petabytes of data without storage becoming the bottleneck. The system's performance benchmarks signal what high-throughput computing resources are available for open scientific projects.

The takeaway

Storage throughput is becoming as critical as processing power for large-scale science, and Aurora's DAOS results show how that balance is being struck in practice. Organizations planning high-performance computing projects can use these benchmarks as a reference point for storage design.