Senior CUDA Kernel Engineer

Ensigncode is hiring a Senior CUDA Kernel Engineer to own kernel implementation on our GPU optimization team: developing and optimizing GEMM, grouped GEMM, SpMM, and other compute-heavy kernels in CUDA C++. We're looking for evidence of measured speedups on real workloads, not just familiarity with CUDA.

Modern GPU-accelerated systems live or die on kernel-level performance. As our Senior CUDA Kernel Engineer, you’ll own the kernel implementation for Ensigncode’s GPU optimization work: designing, writing, and tuning CUDA C++ kernels for GEMM, grouped GEMM, SpMM, and other compute-heavy operations, and knowing when a custom kernel is justified over NVIDIA’s own libraries.

This is not a role for someone who has only read about CUDA. We’re looking for engineers who can show measured, reproducible speedups on real workloads, and who partner naturally with the team’s profiling and numerical specialists to confirm both that a kernel is actually faster and that it’s still correct.

Responsibilities

• Develop and optimize GPU compute kernels using CUDA C++, with a focus on GEMM, grouped GEMM, SpMM, and other compute-heavy workloads
• Use NVIDIA libraries and tools such as cuBLAS, cuBLASLt, CUTLASS, and cuSPARSE; identify when custom kernels are justified
• Optimize thread/block mapping, memory coalescing, shared memory, occupancy, launch overhead, and data movement
• Implement robust kernels with clear correctness tests and maintainable code

Core requirements

• C++ and CUDA; GPU execution model; memory hierarchy; parallel algorithms; kernel-level performance tuning
• Evidence of measured speedups, not only familiarity with CUDA

Team contribution

Owns the kernel implementation. Partners with profiling and numerical specialists to establish why a kernel is slow and whether results remain correct.

Works most closely with