Software Engineer - GPU Kernels

About this role

Baseten seeks a GPU Kernel Engineer to optimize high-performance kernels for AI inference workloads, powering production systems at companies like Cursor and Notion. You'll design and implement cutting-edge CUDA kernels for matrix operations, attention mechanisms, and advanced features like quantization, directly impacting the performance of millions of users.

What you'll do

• Design and implement high-performance GPU kernels for ML operations including matrix multiplications, attention mechanisms, and mixture-of-experts routing
• Write optimized code using CUDA, PTX assembly, and architecture-specific techniques
• Apply performance optimization methods such as memory coalescing, tensor core acceleration, and compute/memory overlap
• Implement cutting-edge features like quantization (FP8/FP4), sparsity, and distributed compute techniques
• Identify and resolve performance bottlenecks using profiling tools like Nsight Systems and Torch Profiler
• Collaborate with research teams to productionize theoretical advancements and contribute to open-source GPU libraries

What they're looking for

• GPU architecture and CUDA programming
• C++ and performance profiling
• Memory hierarchy optimization and bandwidth tuning
• Transformer models and attention mechanism optimization
• GPU kernel libraries (Cutlass, Triton, Thrust, CUB)
• GEMM tuning and distributed multi-GPU compute
• Quantization and numerical precision strategies
• Modern GPU features like tensor cores and async operations