Turbulence Simulations Using GPUs

by

Mahendra Verma

Professor in Physics
JC Bose Fellow, FNA, FASc, FNASc
Indian Institute of Technology Kanpur
INDIA
mkv@iitk.ac.in

Wednesday, May 22

11:00 a.m.

Montgomery Knight Building Rm. 317

About the Seminar:

Graphical Processing Units (GPUs) are superfast with reasonably large
amount of memory. Hence, many top supercomputers are employing them.
To take advantage of these hardware, we have developed two spectral codes,
one based on CUDA and other based on Cupy. On a A100 GPU, the CUDA
and Cupy codes are respectively 3000X and 500X faster than a Rome core. In
addition, these codes scale quite well on multinode and multigpu platforms.
In this talk, I will describe the features and scaling of these codes. I will also
show MHD turbulence results obtained using these codes.
 

About the speaker:

Prof. Mahendra Verma is at the Physics Department of IIT Kanpur. He is
a recipient of J. C. Bose Fellowship, Swarnajayanti Fellowship, Dr. A.P.J.
Abdul Kalam Cray HPC Award, and INSA Teachers Award. He is a fellow
of the three Indian academies. His research interests are turbulence, highperformance
computing, and complex systems. He authored several books,
including “Energy Transfers in Fluid Flows”, “Physics of Buoyant Flows”,
and “Practical Numerical Computing Using Python”.
 

Contact: Prof. P.K Yeung, School of AE (pk.yeung@ae.gatech.edu