Chengming Zhang
University of Houston, TX·Computer Science
Supercomputers built on graphics processing units (GPUs), chips that perform many calculations in parallel, now produce huge amounts of scientific data from simulations. Storing, moving and analyzing that data is slow and expensive.
- Fields
Computer Science
Engineering
- Topics
lossy compression
error-bounded compression
data reduction
- Grants
- 1 known grant, through Jul 2028
What they work on
- Adapt GPU error-bounded lossy compressors such as cuSZ to real scientific workflows, supporting diverse data formats and compression quality targets, with quality checked by tools like QCAT.
- Improve usability of GPU data reduction through high-level language bindings, a command-line interface, and a user interface with integrated visualization, building on LibPressio and HDF5.
- Make state-of-the-art GPU compressors run on NVIDIA, AMD and Intel GPUs using the Kokkos performance-portability framework.
More about the research
Compression shrinks the data. Error-bounded lossy compression goes further: it discards some detail while guaranteeing that every error stays below a limit the scientist sets. However, tools for doing this on GPUs are immature. They are hard to use, often do not fit real analysis needs, and mostly run only on NVIDIA hardware.
Why it matters
Scientists running large simulations on GPU supercomputers can generate more data than they can store or analyze efficiently. A portable, easy-to-use GPU compression toolkit could cut storage and transfer costs and speed up data analysis. The goal is to help researchers across fields work with exascale data on NVIDIA, AMD and Intel systems.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Hanqi Guo
Ohio State University, OH·Computer Science
Develop lossy compression algorithms for scalar and vector field data that preserve topological features, such as critical points, for analysis and visualization in cosmology, climate and fusion simulations.
Yang Zhang
Miami University-Oxford, OH·Computer Science
Characterize the artifacts that error-controlled lossy compressors introduce into scientific data and later analyses, and establish a benchmark for artifact mitigation.
Neranjan Edirisinghe Pathirannehe
Georgia State University, GA·Computer Science
Deploy a campus computing cluster with 12 NVIDIA H100 GPUs, BlueField data processing units, 682 CPU cores and 228 TB of storage to support 11 research projects spanning wildfire management, solar prediction, neuroscience, finance and policy.
The school
University of Houston: 21 labs on PI Pages, $239M a year on research, 374 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.