Intern - Research in Scalable Differentiable Algorithms

NVIDIA is seeking a talented intern to join their team in researching scalable differentiable algorithms. The successful candidate will work on developing GPU-native solvers with adjoint and automatic differentiation techniques for engineering simulations.

The internship will focus on connecting simulations with experimental data for calibration and data assimilation using AI/ML methods. The ideal candidate will have a strong foundation in algorithms related to PDEs, numerical linear algebra, iterative solvers, optimization, or scientific computing.

Responsibilities:

• Prototype GPU-native solvers with adjoint and automatic differentiation techniques for engineering simulations.
• Integrate AI-based methods with simulation algorithms for parameter calibration, inverse problems, and data assimilation using experimental or simulation data.
• Co-design numerical algorithms and differentiation strategies for GPU parallelism, memory hierarchy, mixed precision, reduced synchronization, and distributed multi-GPU execution.
• Evaluate numerical accuracy, convergence, gradient quality, performance, and scalability on workloads in CFD, mechanics, thermal systems, and multiphysics.
• Identify promising methods and help move them toward reusable capabilities in CUDA-X libraries.

Requirements:

• Pursuing a PhD in applied mathematics, scientific computing, computational engineering, computer science, mechanical engineering, aerospace engineering, or equivalent program area.
• Strong foundations in algorithms related to PDEs, numerical linear algebra, iterative solvers, optimization, or scientific computing.
• Research experience with adjoint methods, automatic differentiation, inverse problems, data assimilation, gradient-aware simulation approaches, or a related area.
• Experience writing numerical software in Python and C++, with familiarity with parallel or GPU computing.
• Ability to design computational experiments, evaluate numerical and performance tradeoffs, and communicate technical results clearly.

Benefits:

• Competitive hourly rate: $38-$94 per hour
• Eligibility for intern benefits

How to Stand Out:

• Experience developing, scaling, or analyzing GPU solvers, including multi-GPU or distributed execution.
• Hands-on CUDA experience and use of profiling, benchmarking, or performance analysis to improve numerical algorithms.
• Experience calibrating computational models or assimilating experimental data in scientific or industrial applications.
• Experience with NVIDIA Warp, CUDA-X libraries, PyTorch, JAX, PETSc, Trilinos, MFEM, libCEED, OpenFOAM, or related computational science frameworks.
• Publications, open-source contributions, or deployed research software in numerical methods, scientific computing, differentiable simulation, or computational engineering.