Dmitry Pekurovsky
HPC And AI Networking Technology Architect at NVIDIA
California, United States
Summary
👤
Senior🎓
Top SchoolDmitry Pekurovsky is an HPC and AI networking technology architect with over two decades of experience building and optimizing scalable science software, currently contributing to NVIDIA after a long tenure at the San Diego Supercomputer Center. Trained as a theoretical/computational physicist (PhD, Ohio State), he specializes in parallel algorithms, performance tuning, and enabling large-scale simulations across domains from Lattice QCD to oceanography. He is the author and principal engineer of the widely used open-source P3DFFT library, demonstrating rare end-to-end ownership from algorithm design to documentation and community support. Dmitry combines hands-on optimization on top-tier supercomputers with grant-funded research and user-facing collaborations, and he frequently mentors and trains researchers to push codes to production at scale. Notably, his career blends deep numerical expertise with practical experience in porting and debugging complex scientific workloads for both HPC and emerging AI networking environments.
11 years of coding experience
23 years of employment as a software developer
Bachelor of Science - BS, Physics, Bachelor of Science - BS, Physics at Moscow Institute of Physics and Technology (State University) (MIPT)Doctor of Philosophy - PhD, Physics, Doctor of Philosophy - PhD, Physics at The Ohio State University
Github Skills (34)
fftw10
decomposition10
ifs10
essence10
layout9
simulations9
transform9
extensible8
fft8
chemistry8
e-mail8
simulation8
modular7
parallel7
density-functional-theory7
Programming languages (4)
Github contributions (5)
sdsc/p3dfft.3
Aug 2017 - Jul 2021
P3DFFT++ (a.k.a. P3DFFT v. 3) is a new generation of P3DFFT library that aims to provide a comprehensive framework for simulating multiscale phenomena. It takes the essence of P3DFFT further by creating an extensible, modular structure uniquely adaptable to a greater range of use cases. The users can specify in detail what kind of data layout they would like to use, both in terms of local memory ordering and the processor layout. Just like P3DFFT, P3DFFT++ is a distributed software package, using MPI as the primary method for interprocessor commubnication. It supports 1D, 2D and 3D (to come soon) domain decomposition schemes. As P3DFFT, P3DFFT++ also relies on lower-level libraries, for example FFTW to perform optimized 1D FFTs. Unlike P3DFFT, which was written in Fortran90, P3DFFT++ is written in C++. Interfaces are provided for C and Fortran. To learn about using the code the user is encouraged to study example programs in C++, C and FORTRAN subdirectories. Please e-mail Dmitry Pekurovsky (dmitry@sdsc.edu) for any questions or suggestions. Software contributions are welcome, assuming they follow the main ideas of the framework.
Contributions:5 releases, 92 commits, 2 PRs in 3 years 11 months
casesmemorympifollowsimulation
sdsc/p3dfft
Apr 2015 - Apr 2019
P3DFFT stands for Parallel Three-Dimensional Fast Fourier Transforms. It is a library for large-scale computer simulations on parallel platforms. It implements 3D FFT and related algorithms such as Chebyshev transform (an important class of algorithm for simulations in a wide range of fields). P3DFFT uses 2D, or pencil, decomposition. For more information:
Contributions:5 releases, 56 commits, 3 PRs in 4 years
pencilconvolutionfourierchebyshevplatforms
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.