CUDA Kernel Engineer

Periodic Labs • Menlo Park, CA

90d

About The Position

Periodic Labs is an AI + physical sciences lab focused on building state-of-the-art models for novel scientific discoveries. The company is well-funded and experiencing rapid growth, with a team culture that encourages ownership, problem-solving, and learning new tools and sciences to advance its mission. The role involves developing, integrating, and optimizing CUDA kernels to enhance AI scientific research. This includes integrating CUDA kernels into training, inference, and reinforcement learning systems that operate on thousands of GPUs. The position also entails building tools and supporting frontier-scale experiments to establish Periodic Labs as the leading AI + science lab, with the expectation of releasing kernels as contributions to the open-source AI stack.

Requirements

• Experience writing and optimizing CUDA kernels, including attention, mixture-of-experts, dispatch-and-combine, and others.
• Experience working with the latest generation of Nvidia hardware.
• Experience integrating kernels into state-of-the-art inference frameworks (e.g., vLLM, SGLang) and training frameworks (e.g., Megatron, TorchTitan).

Responsibilities

• Develop, integrate, and optimize CUDA kernels for AI scientific research.
• Integrate CUDA kernels into training, inference, and reinforcement learning systems running on thousands of GPUs.
• Build tools to support frontier-scale experiments.
• Release CUDA kernels as contributions to the open-source AI stack.

What This Job Offers

• Career Level: Mid Level