Triton Compiler/GPU Kernel Performance Engineer

Job Description 

Kernel Performance Architect 

WHAT YOU DO AT AMD CHANGES EVERYTHING 

At AMD, we build the compute engines that power AI, high-performance computing, and next-generation data centers

Our GPU platforms drive breakthroughs across machine learning, scientific computing, and large-scale distributed systems. 

We are looking for a Kernel Performance Architect who can bridge hardware, compiler, runtime, and application layers to define and drive end-to-end performance strategy for AI workloads on AMD GPUs. 

This role is not just about writing fast kernels — it is about understanding why they are fast, predicting performance behavior across architectures, and shaping the abstractions that make performance portable. 

THE ROLE 

The Kernel Performance Architect is responsible for defining, analyzing, and optimizing performance across the full stack — from GPU microarchitecture and compiler behavior to runtime systems and deep learning frameworks. 

You will: 

• Lead performance architecture for key AI workloads 

• Define cross-architecture optimization strategies 

• Diagnose bottlenecks across kernel, memory, compiler, and runtime layers 

• Design performance methodologies and benchmarking frameworks 

• Influence abstraction design to ensure performance portability 

• Collaborate with compiler, runtime, and hardware teams 

This is a high-impact technical leadership role requiring strong performance intuition and systems-level thinking. 

THE PERSON 

You are someone who: 

• Can look at a kernel and predict whether it will be memory-bound or compute-bound before running it 

• Understands how register pressure, occupancy, memory hierarchy, scheduling, and compiler decisions interact 

• Is comfortable reading disassembly and compiler IR 

• Can design experiments to isolate performance bottlenecks 

• Thinks in terms of architecture invariants, not just one GPU generation 

• Can clearly explain performance tradeoffs to both hardware and software engineers 

You combine deep low-level knowledge with broad systems perspective. 

KEY RESPONSIBILITIES 

Cross-Stack Performance Analysis 

• Analyze performance across kernel, compiler, runtime, and framework layers 

• Identify root causes of bottlenecks (memory bandwidth, latency hiding, scheduling limits, interconnect constraints) 

• Build mental and quantitative performance models 

Kernel Architecture & Optimization Strategy 

• Define tiling, scheduling, and memory layout strategies for AI kernels 

• Evaluate tradeoffs between portability and peak performance 

• Guide implementation engineers toward correct optimization directions 

Compiler & Runtime Collaboration 

• Work with compiler teams (LLVM, ROCm) to analyze generated code 

• Influence scheduling, register allocation, and code generation improvements 

• Align kernel design with runtime behavior (streams, synchronization, distributed execution) 

Performance Methodology & Benchmarking 

• Design reproducible benchmarking frameworks