Jack Kosaian
Senior Architect at NVIDIA
Coralville, Iowa, United States
Summary
🤩
Rockstar🎓
Top SchoolJack Kosaian is a Senior Architect at NVIDIA with 11 years of experience specializing in high-performance GPU software and resource-efficient machine learning systems. He holds a PhD in Computer Science from Carnegie Mellon, where his research focused on reliability and efficiency in ML systems, and he translates that rigor into production-grade performance engineering. At NVIDIA he contributes to CUTLASS, improving CUDA kernels, occupancy calculations, and Python interfaces to broaden hardware compatibility. He has augmented the warp framework with batched CUTLASS GEMM support, demonstrating a knack for bridging low-level CUDA optimization with higher-level Python ecosystems. Prior internships at Microsoft, Google, and others underscore a consistent record of shipping scalable systems in both industry and research settings. Based in Coralville, Iowa, he combines academic depth with hands-on contributions to widely used open-source GPU libraries.
10 years of coding experience
8 years of employment as a software developer
Computer Science, Computer Science at University of MichiganBrighton High SchoolDoctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at Carnegie Mellon University
Github Skills (15)
cuda10
gpu-programming10
nvidia10
gemfire10
deeplearning-ai10
deep-learning10
cpp10
gpu10
performance-optimization10
python10
linear-algebra10
tensorrt9
tensor9
c-language8
cprogramming-language8
Programming languages (6)
Github contributions (5)
NVIDIA/warp
Dec 2022 - Dec 2022
A Python framework for high performance GPU simulation and graphics
Role in this project:
Back-end Developer
Back-end DeveloperContributions:7 commits in 9 days
Contributions summary:Jack implemented batched matrix multiplication (GEMM) functionality using the CUTLASS library within the warp framework. This included adding new functions to the `warp/types.py` and `warp/native/cutlass_gemm.cu` files to support batched matrix operations and integrating CUTLASS for optimized performance. The changes also involved modifications to the testing suite to verify the correctness of the implemented functionality. The core contribution focused on enhancing the framework's computational capabilities for GPU-accelerated linear algebra operations.
cudapythongpucomputer-graphicssimulation
NVIDIA/cutlass
Apr 2022 - Dec 2022
CUDA Templates for Linear Algebra Subroutines
Role in this project:Back-end Developer & Performance Engineer
Back-end Developer & Performance EngineerContributions:18 reviews, 7 commits, 14 PRs in 7 months
Contributions summary:Jack primarily contributed to improving the CUTLASS library's functionality and performance. They fixed typos in example code, addressed issues in occupancy calculations for grouped GEMM, and resolved a typo in an example file. Their work also included making the Python interface compatible with non-SM80 targets, indicating efforts to broaden the library's usability across different hardware configurations. The user also made changes to CUDA kernel code, which included fixing an issue that was causing incorrect occupancy calculations.
cudacpplinear-algebranvidiamatrix-multiplication
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.