SGCC

Scientific GPU Compression Cyberinfrastructure

GPU hardware now carries exascale data production, but the cyberinfrastructure around GPU-based scientific data compressors is still maturing. Existing frameworks are not adapted to many data-analysis requirements, there are few user-friendly or off-the-shelf solutions, and support for non-NVIDIA architectures is thin.

SGCC builds a user-friendly, high-performance, portable GPU-accelerated data-reduction cyberinfrastructure for GPU-equipped supercomputing systems, by porting, extending, and optimizing capabilities that already exist.

Thrusts

  • Fit the workflows

    Efficiency and effectiveness in practical scientific data-analysis workflows, with adequate support for diverse data formats and compression-quality targets.

  • Make it usable

    High-level language bindings, a command-line interface, and a user interface integrated with visualization.

  • Make it portable

    State-of-the-art GPU-accelerated compressors running across NVIDIA, AMD, and Intel platforms.

What it builds on

ported, extended, and optimized

  • pSZ/cuSZ family error-bounded lossy compressors

  • GPU lossless encoders the lossless stages

  • QCAT CPU-based compression quality assessment

  • Kokkos multi-backend performance portability

  • LibPressio unified interface for scientific compressors

  • HDF5 the storage layer

Award

Collaborative Research: Elements: SGCC: An Efficient GPU-oriented Data Reduction Cyberinfrastructure for Scientific Data Analysis.

NSF Office of Advanced Cyberinfrastructure, CSSI Elements: one collaborative project across three institutions, $599,988 in total.