GPU Performance Engineer – Neural Reconstruction

This is a fully remote position, open to applicants in California, +4 more states.

📋 Description

• Map out comprehensive neural reconstruction workflows and pinpoint bottlenecks throughout the stages of data loading, initialization, training, rendering, evaluation, and export.

• Enhance CUDA and PyTorch performance for Gaussian Splatting and neural reconstruction tasks, including handling camera/lidar data, multiview batching, large-scene rendering, and memory-sensitive training routes.

• Evaluate GPU performance utilizing tools like Nsight Systems, Nsight Compute, NVTX, PyTorch Profiler, CUDA events, and benchmark dashboards.

• Optimize rendering workloads that are sparse and irregular, which involve tile-level masking/culling, sparse gradients, batching, and multi-GPU execution.

• Convert high-impact Python, NumPy, or PyTorch bottlenecks into efficient implementations using CUDA/C++ or native PyTorch when suitable.

• Ensure that performance enhancements maintain reconstruction quality, numerical accuracy, camera/lidar precision, and production reliability.

• Create reproducible benchmarks, regression tests, and profiling workflows to detect performance and quality regressions at an early stage.

• Work collaboratively with researchers, CUDA engineers, ML engineers, and production teams to transform promising prototypes into maintainable, reviewable, production-quality code.

⛳️ Requirements

• BS, MS, PhD, or equivalent experience in Computer Science, Computer Engineering, Electrical Engineering, Applied Math, Robotics, Computer Vision, Machine Learning, or a related discipline (or equivalent experience) with over 12 years of experience.

• Proficient programming skills in Python and C++!

• Practical experience with PyTorch or a comparable tensor/autograd framework.

• Experience in optimizing GPU-accelerated workloads using CUDA, C++/CUDA extensions, or other relevant GPU programming techniques.

• Hands-on experience with profiling and performance analysis, which includes identifying CPU/GPU bottlenecks, synchronization delays, memory constraints, kernel launch delays, and inefficiencies at the framework level.

• Capability to develop benchmarks and confirm that optimizations uphold correctness, numerical integrity, and user-visible quality.

• Excellent communication skills, particularly the ability to articulate performance trade-offs, risks, and outcomes to both research and engineering collaborators.

🏝️ Benefits

• Equity

• Comprehensive benefits package