GPU Kernel Developer - AI Models

AMD - Advanced Micro Devices

Remote, United Kingdom

THE ROLE:

AMD is looking for a GPU kernel development engineer who is talented in developing high performance kernels for state-of-the-art and upcoming GPU hardware. You will be a member of a core team of incredibly talented industry specialists and will work with the very latest hardware and software technology. 

THE PERSON:

Experienced in GPU kernel development and optimization for AI/HPC applications. Strong technical and analytical skills in GPU computing, hardware architecture, and deep understanding of HIP/CUDA/OpenCL/Triton development. Ability to work as part of a team, deliver to project scope, and communicate to a technical/non-technical audience.

KEY RESPONSIBILITIES:

Develop high performance GPU kernels for key AI operators on AMD GPUs;
Optimize GPU code using structured and disciplined methodology - profiling to identify gaps, roofline-analysis on hardware, identify key set of optimizations, establish uplift and line-of-sight, prototype and develop optimizations;
Support mission-critical workloads in NLP/LLM, Recommendation, Vision and Audio;
Collaborate and interact with system level performance architects, GPU hardware specialists, power/clock tuning teams, performance validation teams, and performance marketing teams to analyze and optimize training and inference for AI;
Work with open-source framework maintainers to understand their requirements – and have your code changes integrated upstream;
Debug, maintain and optimize GPU kernels, understand and drive AI operator performance (GEMM, Attention, Distributed scale-up/out communication, etc.);
Apply your knowledge of software engineering best practices.

PREFERRED EXPERIENCE:

Knowledge of GPU computing (HIP, CUDA, OpenCL, Triton);
Knowledge and experience in optimizing GPU kernels;
Expertise in using profiling, debugging tools;
Core understanding of GPU hardware;
Excellent C/C++/Python programming and software design skills, including debugging, performance analysis, and test design.