Kernel Engineer

About the Role

We are seeking a highly skilled Kernel Engineer to join our team in Bengaluru or Noida. In this role, you will focus on writing and optimizing low-level CUDA and Triton kernels that significantly impact the speed of our training and inference workloads. This position involves highly specialized, close-to-hardware optimization work that few professionals are equipped to perform.

Key Responsibilities

• Develop and fine-tune CUDA and Triton kernels to enhance training and inference performance across various systems and models.
• Analyze GPU workloads to identify and remove bottlenecks related to memory bandwidth and computation that are often overlooked by standard frameworks.
• Create and implement quantization and mixed-precision kernels (FP8/INT4) to reduce inference costs substantially.
• Collaborate closely with runtime engineering teams to integrate kernel-level improvements into production environments swiftly.
• Monitor advancements in GPU architectures and adapt kernels to leverage new hardware capabilities effectively.

Candidate Profile

• At least 8 years of experience in systems engineering, GPU programming, or performance optimization.
• Extensive practical experience in writing CUDA or Triton kernels.
• Deep knowledge of GPU microarchitecture and expertise in optimizing for memory bandwidth constraints.
• Proficient in C++ with a strong focus on system-level programming and performance considerations.
• Experienced in profiling and debugging down to individual GPU instructions.

Educational Qualifications

A B.Tech or M.Tech degree in Computer Science, Electrical Engineering, or Physics from a top-tier institution.

Desirable Skills

• Experience with PTX or SASS level optimizations.
• Prior experience working at semiconductor companies, EDA vendors, or teams developing GPU or accelerator software.
• Familiarity with quantization or low-precision inference methodologies.