Jiri Kraus
Principal Developer Technology Engineer at NVIDIA
Bonn, North Rhine-Westphalia, Germany
Summary
🤩
Rockstar🎓
Top SchoolJiri Kraus is a Principal Developer Technology Engineer at NVIDIA with 13 years of experience applying numerical methods, parallel computing and GPU performance engineering to real-world HPC and ML workloads. He combines a strong academic background in mathematics with hands-on optimization of multi-GPU codes, kernel-level tuning and scalable data structures, contributing to high-profile open-source projects such as RAPIDS (cuML, cuDF) and NVIDIA’s Parallel Forall examples. His work spans algorithmic improvements (hash joins, multi-pass TLB-aware approaches) to practical usability enhancements and examples that help onboard others to GPU ML. At NVIDIA he progressed from Developer Technology Engineer to principal level, advising customers and improving production-grade libraries for machine learning and dataframes. Known for digging into low-level bottlenecks, he also brings an uncommon mix of actuarial attention to correctness from an early industry role and the research rigor honed at Fraunhofer.
13 years of coding experience
4 years of employment as a software developer
Diplom, Mathematik, Diplom, Mathematik at Universität KölnUniversität Hamburg / University of Hamburg
German, English
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (25)
c-language10
gpu-programming10
mpi10
cudf10
machine-learning10
machine-learning-algorithms10
kernel10
nvidia10
parallel-computing10
gpu10
performance-optimization10
hashtable10
cuda10
hash10
cprogramming-language10
Programming languages (6)
Github contributions (5)
NVIDIA/multi-gpu-programming-models
May 2018 - Oct 2022
Examples demonstrating available options to program multiple GPUs in a single node or a cluster
Role in this project:
Back-end Developer & Performance Engineer
Back-end Developer & Performance EngineerContributions:4 reviews, 53 commits, 7 PRs in 4 years 5 months
Contributions summary:Jiri primarily focused on optimizing the performance of multi-GPU programming examples within the repository. Their contributions include adding optimization flags for atomic operations and integrating CUB for faster residual reductions in the Jacobi solver implementations. The user also improved the norm calculation in the multi-threaded p2p variant and reduced the number of barriers for norm calculations. Furthermore, the user increased block sizes to reduce required atomics, demonstrating a focus on kernel-level optimization.
single-nodecudagpu-accelerationmpiinfiniband
NVIDIA-developer-blog/code-samples
Mar 2013 - Apr 2018
Source code examples from the Parallel Forall Blog
Role in this project:Back-end Developer
Back-end DeveloperContributions:17 commits, 1 PR, 4 pushes in 5 years 2 months
Contributions summary:Jiri contributed source code examples from the Parallel Forall Blog, likely focusing on CUDA-aware MPI. They made changes to include CUDA 4.2 compatibility and added NVTX examples, suggesting an interest in performance analysis and profiling using NVIDIA tools. Additional modifications were made to include NVML examples for monitoring and managing GPU behavior.
gpucudaparallel
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.