GPU Performance Architect - Shape Next-Gen AI GPUs

Socket.dev; Santa Clara (CA); On-site; USD 180,000 - 260,000; Full time

Job summary

AMD's Radeon Technologies Group is seeking a GPU Performance Architect in Santa Clara, CA. You will analyze workloads, identify bottlenecks in graphics cores, and work with architects to drive PPA improvements on next-gen GPUs. Engage with teams across sites to deliver high-impact results.

Ideal candidates have deep experience in GPU architecture, AI/ML algorithms, and GPU performance tooling. Strong C/C++ and Python scripting skills, plus DirectX/Vulkan/CUDA/HIP familiarity, are essential.

Qualifications

• Undergrad degree in EE/CS/Computer Architecture is required.
• Master's/PhD with emphasis in EE/CS preferred.
• Experience across architecture, performance analysis and AI/ML/graphics algorithms.

Responsibilities

• Analyze workloads and game/AI/ML performance across current and future GPUs.
• Identify bottlenecks in graphics cores and SoCs and propose solutions.
• Collaborate with architects to drive RTG analysis improvements.
• Develop and run RTL and high-level simulation models to estimate gains.
• Drive initiatives to improve RTG tooling and workflows.
• Work with teams across multiple sites to deliver results.

Skills

GPU architecture & compilers

LLM kernel development

C/C++/Python scripting

Graphics/Compute APIs

AI/ML/graphics algorithms

GFX profiling tools

Education

Bachelor's degree in Electrical Engineering/Computer Architecture/CS

Master's or PhD preferred

Job description

AMD's Radeon Technologies Group is seeking a GPU Performance Architect in Santa Clara, CA. You will analyze workloads, identify bottlenecks in graphics cores, and work with architects to drive PPA improvements on next-gen GPUs. Engage with teams across sites to deliver high-impact results.

Ideal candidates have deep experience in GPU architecture, AI/ML algorithms, and GPU performance tooling. Strong C/C++ and Python scripting skills, plus DirectX/Vulkan/CUDA/HIP familiarity, are essential.