NVIDIA is seeking a CUDA Math Libraries Intern for 2026 to contribute to the development of accelerated mathematical and data processing libraries. You'll join a team building foundational tools like cuSparse, cuBLAS, cuSOLVER, cuFFT, and DALI that power high-performance computing and deep learning stacks.

What You'll Do

Collaborate with team members to understand software use cases and requirements.
Analyze the performance of GPU or CPU implementations and find opportunities for improvements.
Prototype and develop algorithms for single node and multi GPU clusters.
Implement new algorithms, define APIs, analyze performance, and find solutions for difficult numerical corner cases.
Work on extending capabilities of existing libraries and building new libraries for AI and HPC applications.

What We're Looking For

Studying towards a MS or PhD degree in Computational Science, Computer Science, Applied Mathematics, Engineering, or a related field.
Programming skills in C/C++ and Python.
Parallel or GPU programming experience such as AVX, NEON, OpenMP, MPI, SHMEM, CUDA or OpenCL.

Nice to Have

Exposure to floating-point arithmetic and numerical error analysis.
Knowledge of GPU/CPU and network hardware architecture.
Understanding of composability and fusions, compilers, and implementation of programming languages.
Experience implementing sparse or dense linear algebra algorithms.
Experience with domain-specific language design and compiler optimizations, in particular sparse compilers like MLIR or TACO.