Senior GPU Kernel Engineer for AI & LLMs

Advanced Micro Devices, Inc.; San Jose (CA); On-site; USD 150,000 - 230,000; Full time

Job summary

Advanced Micro Devices, Inc. seeks a senior technical contributor to design, develop, and optimize high-performance GPU kernels for LLMs and ML workloads.

You will analyze bottlenecks and craft architecture-aware optimizations in collaboration with GPU architecture, compiler, library, and framework teams. You will lead initiatives with strong C++/Python skills, GPU compute knowledge, and experience with HIP/CUDA.

Qualifications

• Strong programming skills in C++ and Python with a track record of performance-focused development.
• Deep understanding of GPU compute, memory hierarchy, synchronization, and bottlenecks.
• Experience designing and optimizing GPU kernels using HIP, CUDA, C/C++, or similar.
• Proactive, self-driven with the ability to lead tasks to completion.
• Good collaboration across teams and strong problem-solving skills.

Responsibilities

• Architect and optimize AI workloads for AMD GPU architectures.
• Develop high-performance GPU kernels with a focus on hardware utilization.
• Drive performance analysis across compute, memory bandwidth, and cache hierarchies.
• Advance compiler and code-generation techniques using LLVM/MLIR.
• Innovate with AI-assisted development for kernel generation and benchmarking.
• Collaborate with model and infrastructure teams to improve end-to-end AI system performance.

Skills

C++ & Python

GPU kernel development

Performance analysis

Self-starter

Education

BS/MS in CS/CE

Tools

HIP

CUDA

LLVM/MLIR

Job description

Advanced Micro Devices, Inc. seeks a senior technical contributor to design, develop, and optimize high-performance GPU kernels for LLMs and ML workloads.

You will analyze bottlenecks and craft architecture-aware optimizations in collaboration with GPU architecture, compiler, library, and framework teams. You will lead initiatives with strong C++/Python skills, GPU compute knowledge, and experience with HIP/CUDA.