Software Engineer - C++ GPU Performance
Optimizes GPU-based algorithms for self-driving systems by instrumenting, monitoring, analyzing performance hotspots, and porting serial code to GPU. Requires 3+ years experience, strong C++/CUDA skills, BS in CS, and expertise in recent GPU architectures.
About the job
Responsibilities
- Build real-time instrumentation for performance monitoring (CPU, GPU, latency, memory) and develop offline benchmarking frameworks, tools, and scripts to evaluate & analyze performance at scale in CI/vehicle, and establish budgets for next-gen architectures.
- Analyze performance metrics to identify GPU hotspots and root causes, and propose and co-implement actionable solutions with component teams.
- Support teams on bringing serial algorithms to the GPU to maximize compute utilization and improve overall latency.
- Work as part of the Core team to design a middleware framework that promotes by default efficient and performant code development by maximizing CPU and GPU.
Qualifications
- BS in computer science or related field and 3+ years of experience.
- Strong knowledge of CUDA as applied to recent GPU microarchitectures (e.g., Ampere, Blackwell) and experience debugging/optimizing GPU kernels using tools like Nsight.
- Strong knowledge of C++ and experience in large code bases, comfortable in Linux development environments.
- Experience in development, debugging, and profiling of complex multiprocess systems (e.g., robotic systems, game engines).
Bonus Qualifications
- Experience with GPU kernel development in a real-time environment, including PTX-level programming, CPU SIMD instructions (e.g., AVX intrinsics), and custom CUDA layers with frameworks like TensorRT & XLA.
- Hands-on work with ML model optimization (post-training quantization, layer pruning, etc) or hand-tuning GPU kernels (in OpenGL, CUDA, RocM or similar).
- Proficiency with SQL, DataBricks, Looker, or other business intelligence tools.
Skills
C++, CUDA, Nsight, GPU, Linux, TensorRT, Xla, Avx, Opengl, Rocm