| The Advanced Photon Source (APS) at Argonne organizes an ongoing seminar series, the APS Scientific Computation Seminar Series, focused on various aspects of computation for synchrotron science; talks are related to data analysis, reconstruction, theory, simulation, optimization, machine learning, automation, and visualization. The seminar provides an opportunity to present and learn about state-of-the-art software and the application of computational, AI/ML and mathematical techniques to synchrotron science. |
| Title: | Integrating ORNL’s HPC and Neutron Facilities with a Performance-Portable CPU/GPU Ecosystem | | Presenter(s): | Steven E. Hahn, Software Scientist, Computer Science and Mathematics Division, Oak Ridge National Laboratory | | Date: | May 19, 2025 | | Time: | 1:00 PM (Central Daylight Savings Time) | | Location: | Join ZoomGov Meeting
https://argonne.zoomgov.com/j/1601444470?pwd=N1phbHZVdCtmcVR5cGh0c1Zhc0…
Meeting ID: 160 144 4470
Passcode: 937918
One tap mobile
+16692545252,,1601444470# US (San Jose) +16468287666,,1601444470# US (New York) Dial by your location
+1 669 254 5252 US (San Jose)
+1 646 828 7666 US (New York)
+1 646 964 1167 US (US Spanish Line) +1 669 216 1590 US (San Jose)
+1 415 449 4000 US (US Spanish Line) +1 551 285 1373 US
Meeting ID: 160 144 4470
Find your local number: https://argonne.zoomgov.com/u/af2crdvQy
| | Abstract: | We explore the development of a performance-portable CPU/GPU ecosystem to integrate the Oak Ridge Leadership Computing Facility (OLCF) and the Spallation Neutron Source (SNS), both of which are housed at Oak Ridge National Laboratory. We select a data reduction workflow use-case to obtain the differential scattering cross-section from data collected by SNS's CORELLI and TOPAZ instruments. Inconvenient execution times from complex data processing techniques and large volumes of data discourages users from utilizing all experimental data. The current CPU-only production implementation using the Garnet Python multiprocess package based on the C++ Mantid framework is compared against our proposed CPU/GPU implementation that uses the LLVM-based Julia scientific language and the JACC.jl performance portability package. To understand and address performance challenges, two proxy apps were developed: (i) an app for extracting relevant Mantid kernels (MDNorm) in C++ and (ii) the Julia MiniVATES.jl miniapp. We introduce algorithmic improvements: (i) parallelization strategies across multiple GPU nodes, (ii) reformulating key calculations into a tall-skinny matrix multiplication kernel, and (iii) exploring intermediate data layout representations for optimizing reading and processing times. Performance results from NVIDIA A100, H200 and AMD MI100 GPUs and AMD EPYC and NVIDIA Grace CPUs will be presented. These computational experiments show 10x+ speed ups and provide insights for future generations of data reduction software that can take advantage of developments in productivity and performance portability, making high-performance computing more accessible for an integrated research infrastructure across DOE's experimental and computational facilities. | | Bio: | Dr. Steven Hahn is a Software Scientist in the Computer Science and Mathematics Division at Oak Ridge National Laboratory (ORNL). He has been contributing to many research software projects for power grid, neutron scattering, materials science, and plasma physics. Prior to his current role, he was a Data Analysis Software Engineer in the Neutron Data Analysis and Visualization Division and a Postdoctoral Research Associate in the Quantum Condensed Matter Division at ORNL. He received his PhD in Condensed Matter Physics from Iowa State University in 2012. |
|
