Principal Engineer, CUDA UMD - GPU Kernel Scheduling

NVIDIA•Santa Clara, CA

2d•$272,000 - $431,250

About The Position

NVIDIA is seeking a seasoned system software engineer to work on the CUDA Driver, a core component of its platform for accelerating general-purpose computation on GPUs. The role contributes features and improvements that help realize NVIDIA hardware’s potential across workloads including deep learning, scientific computation, data science, self-driving cars, video games, and virtual reality. The engineer will collaborate with colleagues across NVIDIA to shape CUDA’s future direction and deliver an effective compute platform. The position focuses on GPU kernel scheduling, including exploring how CUDA Graphs can improve the scheduling efficiency and speed of AI/ML workloads. The engineer will help extend CUDA programming models and functionality, including CUDA Graphs, and will develop code for multiple operating systems.

Requirements

BS or MS degree in Computer Science, Electrical Engineering, or a related field, or equivalent experience.
Strong C and C++ programming skills.
Minimum of 15+ years of related development experience; the posting notes that positions for varying experience levels are open.
Experience driving projects across multiple teams.
Experience working with large codebases.
Background with operating system interfaces for threads, process control, and virtual memory.
Experience writing and debugging multithreaded programs.
Good written communication and presentation skills.

Nice To Haves

Prior experience with parallel computing, preferably writing CUDA programs or libraries that use CUDA.
Understanding of system-level architecture, such as interconnects, memory hierarchy, interrupts, and memory-mapped I/O.
Knowledge of memory coherence and consistency models.
Background with kernel-mode development.
Experience with Linux systems software development and with maintaining and extending programming models or higher-level language support for similar environments.

Responsibilities

Evangelize, architect, and implement new features.
Coordinate and drive development efforts across multiple teams.
Help define forward-looking improvements to CUDA APIs and the programming model.
Extend CUDA programming models and functionality, such as CUDA Graphs.
Explore ways to use Graphs to improve the efficiency and speed of AI/ML workload scheduling on GPUs.
Write effective, maintainable, and well-tested code.
Develop code for multiple operating systems.