Senior Applied Research Scientist – GPU Native Numerical Algorithms

Description

The company builds software, systems and research platforms that let scientists and engineers solve problems that were previously out of reach.

The team is a computational engineering applied research group that designs GPU-native numerical methods to make engineering simulation faster, more reliable and easier to use, and that supplies the numerical base for AI-native engineering algorithms.

The work is not a straight port of CPU methods. The aim is to redesign methods around massive parallelism, memory hierarchy, reduced synchronization, mixed precision, tensor cores and multi-GPU systems.

Responsibilities

• Explore solver algorithms, build research prototypes and compare approaches on representative workloads.
• Design new numerical algorithms whose structure fits modern the company GPUs, for implicit and explicit engineering simulation.
• Build linear and nonlinear solvers, including Newton-Krylov, multigrid and AMG, domain decomposition, matrix-free methods, mixed precision, sparse iterative and direct methods, and preconditioners.
• Decide when a CPU-oriented method should be reworked or replaced for GPUs, covering synchronization-avoiding Krylov methods, GPU multigrid and domain decomposition, matrix-free implicit methods, mixed-precision algorithms and sparse direct and iterative hybrids.
• Test algorithms on mechanics, contact, thermal-fluid, electromagnetics, semiconductor process and device, EDA, multiphysics and other CAE workloads.
• Work with CUDA-X, Warp, solver engineering, the company Research, universities and industry partners to take useful research to production the company software.
• Help set the long-term applied research roadmap for GPU numerical methods and their path toward AI-native computational engineering.

Requirements

• PhD or equivalent experience in computational mechanics, applied mathematics, scientific computing, computer science, aerospace, mechanical or civil engineering, or a related technical field.
• 5+ years of relevant work or research experience.
• Research or engineering experience with PDE discretization, finite element, finite volume, discontinuous Galerkin, nonlinear solvers, sparse linear algebra, preconditioning or high-performance computing.
• Experience writing numerical software in C++ and Python, and developing or optimizing CUDA or GPU code.
• Experience using profiling, benchmarking, numerical validation or performance analysis to improve algorithms on GPU or multi-GPU systems.
• Ability to explain technical tradeoffs clearly and work across research, engineering, product and partner teams.
• Ways to stand out: experience with implicit structural dynamics, nonlinear mechanics, contact, CFD, electromagnetics, multiphysics, semiconductor simulation, EDA, CAE, or CAD-connected engineering workflows.
• Ways to stand out: contributions to or hands-on use of PETSc, Trilinos, MFEM, libCEED, OpenFOAM, the company Warp, CUDA-X, cuSPARSE, cuSOLVER or related computational science frameworks.
• Ways to stand out: experience with industrial simulation, EDA, semiconductor, CAE or CAD ecosystems such as Ansys, Abaqus, LS-DYNA, Siemens Simcenter, Dassault SIMULIA, Altair, Cadence, Synopsys, COMSOL, MathWorks, or comparable internal solver and design platforms.
• Ways to stand out: experience with distributed solvers using MPI, NCCL or asynchronous methods, or with performance analysis on GPU clusters.
• Ways to stand out: publications, patents, open-source work or deployed software in computational science venues or communities such as SC, SIAM CSE, SIAM SISC, CMAME, IJNME, JCP, AIAA, USNCCM, WCCM or related areas.