Davide Rossetti
Principal Software Engineer at Nvidia
Redwood City, California, United States
Summary
🤩
Rockstar🎓
Top SchoolDavide Rossetti is a Principal Software Engineer with over two decades of deep expertise in HW/SW co-design for high-performance computing, currently at NVIDIA after a long research career at INFN. He designs low-latency PCIe-based interconnects and kernel drivers that enable GPU peer-to-peer communication and RDMA, and has contributed performance-critical kernel work (SSE4.1/AVX) to NVIDIA’s gdrcopy project. Comfortable across device drivers, custom processor back-ends, and numerical libraries, he blends systems-level engineering with practical GPU/CUDA optimizations to squeeze maximum throughput from many-core platforms. His background includes building a production 3D Torus GPU-aware interconnect and adding native complex-type support to GCC for custom supercomputers—an uncommon mix of compiler, OS and hardware-level accomplishments. Based in Redwood City, he pairs academic rigor (Physics, Sapienza) with hands-on performance engineering for cutting-edge HPC and GPU systems.
11 years of coding experience
17 years of employment as a software developer
110/110 cum laude Physics, 110/110 cum laude Physics at Sapienza Università di Roma58/60 standard course, 58/60 standard course at Liceo Scientifico A.Righi - Roma
Italian
Stackoverflow
11reputation
258reached
1answer
0questions
Github Skills (10)
c1710
avx10
c1110
sse10
performance-optimization10
linux9
nvidia9
cuda-driver6
cuda6
synchronization6
Programming languages (5)
Github contributions (5)
NVIDIA/gdrcopy
Dec 2014 - Dec 2022
Role in this project:
Back-end Developer & Performance Engineer
Back-end Developer & Performance EngineerContributions:5 releases, 117 reviews, 211 commits in 8 years 1 month
Contributions summary:Davide primarily contributed to the kernel-mode driver component of the gdrcopy library, which focuses on high-performance GPU memory operations. They added SSE4.1 and AVX-optimized implementations for memory copy functions, particularly for efficient data transfer from the GPU's BAR. The user also made several optimizations, including improved alignment checks and the use of write-combining memory for performance gains. Moreover, the user implemented features to improve the stability and reliability of the driver.
cudamemorynvidiagpulinux
gpudirect/libmlx5
Dec 2016 - Jul 2018
Contributions:27 commits, 1 PR, 13 pushes in 1 year 7 months
user-spacemellanoxuser-space-driver
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.