Location: Lemont, IL
Cap-Exempt H-1B Position — No lottery required
The Mathematics and Computer Science (MCS) division at Argonne is seeking a Predoctoral Appointee to work on enabling HDF5 to support data compression on GPU. HDF5 provides built-in compression support through the H5Z filter mechanism, which allows developers to integrate compression algorithms so that data can be automatically compressed and decompressed during I/O operations. However, the current H5Z filter framework is designed with CPU-based compressors in mind and does not effectively support GPU-based compression. In this project, we aim to integrate GPU compressors into the HDF5 ecosystem. This may involve developing GPU-enabled H5Z filters, potentially leveraging the Virtual Object Layer (VOL), as well as exploring hybrid approaches that combine GPU compressors with the existing H5Z filter mechanism. The resulting solution will be designed to be extensible across different GPU compressors and compatible with frameworks such as libpressio. We will conduct a comprehensive performance evaluation of the proposed approach, with the goal of producing one to two high-quality research publications. **Position Requirements** - A recently completed Master’s degree in Computer Science, Computational Science, Applied Mathematics, or a related field. and 0+ years of experience - Strong programming skills in C/C++ (experience with Go/Python is a plus) - Solid understanding of parallel computing concepts and performance optimization - Experience with parallel programming frameworks, including MPI, OpenMP, and CUDA - Familiarity with GPU architectures (e.g., NVIDIA A100, V100) and heterogeneous computing - Experience with HPC environments, including job schedulers (e.g., SLURM), cluster-based development and benchmarking - Knowledge of numerical algorithms and scientific computing (e.g., iterative methods, FFT/DFT, matrix operations) - Ability to model Argonne's core values of impact, safety, respect, integrity and teamwork **Job Family** Temporary **Job Profile** Predoctor