Software Engineer - GPU Kernels

What you'll do

• Build high-performance kernels for ML operations such as matrix multiplication, attention, and mixture-of-experts routing.
• Optimize GPU performance using architecture-specific techniques and profiling tools; implement quantization, sparsity, and compute/communication overlap.
• Collaborate with research teams to productionize advances and contribute to GPU libraries.

What they ask for

• Strong understanding of GPU architecture, memory hierarchy, thread/block/grid organization, and synchronization.
• Proficiency in C++ and GPU performance profiling tools.
• Knowledge of the CUDA C++ API, memory access and bandwidth optimization, numerical precision and quantization, and modern GPU features.

Nice to have

• Experience with Transformer models and attention optimization, such as Flash Attention.
• Familiarity with GPU kernel libraries such as Cutlass, Triton, Thrust, or CUB.

Pay

$180K – $360K USD base salary

Tools

CUDA, PTX assembly, C++, Nsight Systems, Nsight Compute, Torch Profiler, Cutlass, Triton

AI focus

Optimize GPU kernels and inference performance for machine learning workloads, including quantization, sparsity, and attention mechanisms.