Nvidia Senior Computer Vision System Performance Engineer 

United States, Texas 

291286680

What you'll be doing:

• Develop, profile and optimize data-center and edge computer vision workloads for efficiency, latency, and throughput (Python).

• Implement and improve computer vision and image processing algorithms using CUDA.

• Upstream performance improvements to SDKs and libraries across NVIDIA to deliver accelerated computer vision at scale.

• Influence software architecture, validation strategy and technical roadmaps to ensure outstanding performance.

• Promote high-performance computer vision across NVIDIA teams and functions (Engineering, Product Management, Marketing, and more).

What we need to see:

• Master's of Science in Computer Science or Electrical engineering or equivalent experience.

• 8 years of practical experience.

• Excellent software engineering fundamentals (source control, CI/CD, testing/validation, packaging, containerization, release). Proven track record developing, testing and releasing production-grade, complex software.

• Proficiency with Python, CUDA and C++.

• Strong fundamentals with multi-threaded, multi-process and distributed software development. Expertise defining and driving performance metrics through profiling and benchmarking.

• Experience developing performance-critical data center and cloud applications (REST APIs, gRPC).

• Excellent written, visual, and verbal communication to present performance challenges, tradeoffs, and architectural alternatives.

• Curiosity and drive to learn new technologies and partner across teams and functions.

Ways to Stand Out from the Crowd:

• Expertise in classical, non-ML computer vision

• Expertise in ML computer vision (VLMs, Vision Transformers, Diffusion models) and its software ecosystem: PyTorch, HuggingFace, vLLM

• Grounding in mathematical fundamentals such as linear algebra, numerical methods, statistics, and exploratory data analysis.

• History of creativity and innovation around performance in multiple problem domains.