Search Constraints
1 entry found
Number of results to display per page
View results as:
Search Results
-
- Creator:
- An, Yifu
- Description:
- We have ported our MHD code, BATSRUS ( https://github.com/SWMFsoftware/BATSRUS), to the GPU. This dataset contains the input parameters and raw timing results for the Paper (being revised). To reproduce the results on the supercomputers Frontera and Pleiades, the user need to load modules nvhpc 20.7 for compiling on frontera, nvhpc 24.3-nompi + mpi-hpe/mpt for compiling on Pleiades. and Abstract: BATSRUS, our state-of-the-art extended magnetohydrodynamic code, is the most used and one of the most resource-consuming model in the Space Weather Modeling Framework. It is ever our objective to improve its efficiency and speed with emerging techniques, such as GPU acceleration. To utilize the GPU nodes on modern supercomputer, we port BATSRUS to GPUs with the OpenACC API. Porting the code to a single GPU requires to rewrite and optimize the most used functionalities of the original code into a new solver, which accounts for around 1% of the entire program. To port it to multiple GPUs, we implement a new message passing algorithm to support its unique block adaptive grid feature. We conduct weak scaling tests on as many as 64 GPUs on two supercomputers, and find good performance for up to 16 GPUs. The program is able to retain 65-80% parallel efficiency within one node (up to 4 GPUs), and 40-60% efficiency for 16 GPUs. Applications on more GPUs have reduced efficiency but remain operationally meaningful. Profiling shows that the loss of efficiency is due to the communication latency between supercomputer nodes. We also demonstrate our ability to run representative magnetosphere simulations on GPUs. The performance for a single V100 GPU is about the same as 140 Intel Xeon E5-2680v4 ''Broadwell'' CPU cores, and it runs about twice as fast as real time. The code can run 4 times faster than real time on 4 V100 or 16 RTX 5000 GPUs.
- Keyword:
- BATSRUS, GPU, and MHD simulation
- Citation to related publication:
- An, Y., Chen, Y., Zhou, H. and Toth, G. (2024). BATSRUS GPU: Faster than Real Time Magnetospheric Simulations with a Block Adaptive Grid Code. Being revised.
- Discipline:
- Engineering