Senior Software Engineer - GPU Kernel Authoring & Optimization

About the role

The Senior Engineer for CoreWeave's Benchmarking & Performance team will write, profile, and tune GPU kernels, focusing on optimizing the critical path of large-scale model serving. Responsibilities include authoring, profiling, and optimizing CUDA kernels, optimizing for hardware, implementing and maintaining benchmarking workflows, and leading design reviews.

Responsibilities

• Write, profile, and tune CUDA kernels for GEMMs, attention, MoE routing, quantization, KV-cache, and fused epilogues.
• Optimize for hardware: exploit tensor cores and tune occupancy, memory coalescing, shared-memory/register usage, and overlap of compute with data movement.
• Prototype and ship kernels quickly using kernel-authoring DSLs and compilers, ensuring performance.
• Benchmark rigorously: build reproducible microbenchmarks and roofline analyses, validating kernel-level wins across model-serving stacks (vLLM, TensorRT-LLM, llm-d, SGLang).
• Implement and maintain benchmarking workflows for end-to-end MLPerf Inference (and Training) runs, including workload setup, cluster configuration, runbooks, and result validation.
• Lead design reviews and drive architecture within the team; decompose multi-service work into clear milestones.
• Mentor junior engineers; review cross-team designs and elevate coding/testing standards.
• Help ensure reproducible, well-documented benchmarking and kernel-optimization processes.

Requirements

• 5+ years of experience building high-performance computing, GPU/accelerator software, or performance-critical systems.
• Hands-on CUDA experience, writing and optimizing custom kernels, fluent with CUDA programming and memory model.
• Deep understanding of GPU architecture and performance: tensor cores, warp/occupancy tuning, memory hierarchy, NVLink/PCIe, and profiling with Nsight Compute/Systems.
• Strong coding in C++ and Python; comfortable with low-level, performance-sensitive code.
• Familiarity with model-serving stacks (vLLM, TensorRT-LLM, llm-d, SGLang) and the kernels that dominate their inference cost.
• Strong communicator, comfortable collaborating with cross-functional teams and external partners.

Preferred

• Triton or Mojo for authoring custom GPU kernels.
• CuTe DSL for Python-based kernel authoring on NVIDIA GPUs.
• JAX and its Pallas kernel language for authoring kernels on GPU/TPU.
• HIP / ROCm and AMD GPU experience.
• NCCL and collective-communication performance.
• Experience with alternative accelerators such as Google TPUs and Meta's MTIA.
• Familiarity with kernel-authoring DSLs and nano-compilers such as KNYFE and its Block DSL.
• Experience with Kubernetes at production scale.
• Experience with SUNK (Slurm on Kubernetes) / Slurm for scheduling large GPU jobs.
• Experience running MLPerf submissions or similar large-scale audited benchmarks.
• Contributions to OSS projects such as vLLM, SGLang, PyTorch, Triton, or CUTLASS.

Who You Are

Apply now if you are excited about squeezing every last microsecond out of GPU kernels and delivering reliable model serving. We are looking for someone who can bring their own diversified experiences to our teams.