CUDA Kernel Engineer

📋 Description

• • Join a pioneering AI startup, recognized as a Top 10 GenAI company, founded by esteemed researchers from MIT CSAIL, and contribute to the core GPU performance layer of cutting-edge AI systems.
• • This is a unique opportunity to work on the foundational elements that power large-scale, high-throughput AI solutions for Fortune 500 clients, directly impacting the efficiency and scalability of mission-critical applications.
• • You will be instrumental in designing, implementing, and meticulously optimizing custom CUDA kernels from the ground up, specifically tailored for NVIDIA GPUs.
• • The primary focus will be on maximizing GPU performance metrics, including achieving peak occupancy, enhancing memory throughput, and ensuring optimal warp execution efficiency.
• • Engage in in-depth performance analysis of GPU workloads, leveraging industry-standard profiling tools such as NVIDIA Nsight Compute, Nsight Systems, nvprof, and CUDA‐MEMCHECK to identify and diagnose performance bottlenecks.
• • Take ownership of analyzing and systematically eliminating performance limitations, including but not limited to warp divergence, uncoalesced memory access patterns, excessive register pressure, and inefficiencies in PCIe transfer overhead.
• • Play a key role in refining and optimizing GPU memory pipelines, encompassing global, shared, L2, and texture memory, ensuring that memory access patterns are coalesced for maximum efficiency.
• • Collaborate dynamically with cross-functional teams, including AI systems engineers, model acceleration specialists, and backend distributed systems developers, to integrate and enhance GPU performance across the entire AI stack.
• • Contribute significantly to strategic GPU architecture decisions, the development of robust kernel libraries, and the establishment and enforcement of internal best practices in performance engineering.
• • This role demands a deep, practical understanding of NVIDIA GPU architecture, including its intricate memory hierarchy, warp-level execution models, and comprehensive profiling workflows.
• • You will be expected to go beyond simply utilizing existing libraries and demonstrate a proven track record of building and optimizing CUDA kernels from scratch.
• • Develop a nuanced understanding of CUDA's execution model, including threads, warps, blocks, and grids, and how they interact with the GPU's memory hierarchy.
• • Gain expertise in diagnosing and mitigating performance issues related to memory coalescing and warp divergence, understanding how to detect, analyze, and resolve these common bottlenecks.
• • Become proficient in identifying and addressing PCIe bottlenecks, optimizing data transfers between the host and device through techniques like pinned memory, asynchronous streams, efficient batching, and overlapping computation with communication.
• • Work with modern C++ and CUDA runtime APIs, utilizing advanced GPU debugging and profiling tools to ensure code correctness and performance.
• • This position offers a direct line of sight into the impact of your work, as the AI solutions you help optimize are deployed by leading Fortune 500 companies.
• • You will be part of an environment that fosters innovation and growth, with a strong research pedigree and a history of successful exits by its alumni, including acquisitions by Databricks, Nvidia, and CoreWeave.
• • The company has secured significant funding and is poised for further growth, offering a stable yet dynamic environment for career advancement.
• • This role provides an exceptional opportunity for career growth and influence, allowing you to lead AI initiatives, refine critical performance pipelines, and make a tangible impact on production AI systems at a massive scale.
• • Embrace a culture that values autonomy and collaboration, where you can own critical systems while working alongside a team of world-class engineers.
• • Tackle some of the most challenging and aspirational GPU/AI performance problems in the industry, pushing the boundaries of what's possible in AI computation.
• • The company is committed to providing equal employment opportunities and fostering a fair and inclusive hiring process, processing personal data solely for recruitment purposes in accordance with privacy laws.