Senior Triton GPU Kernel Engineer for AI Workloads

Socket.dev; San Jose (CA); Hybrid; USD 180,000 - 300,000; Full time

Benefits offered by this job

AMD benefits at a glance

Job summary

AMD is seeking a highly skilled software engineer to design, implement, and optimize Triton/Gluon kernels for ML workloads on AMD Instinct accelerators. You will collaborate with research, compiler, and hardware teams to co-design high-performance solutions and push the limits of GPU performance.

You will influence end-to-end performance from kernel coding to runtime integration, contribute to open-source Triton and ROCm ecosystems, and mentor other engineers in SIMT programming and memory

Qualifications

• Deep experience in GPU kernel development and performance optimization for AI/ML workloads.
• Strong background in SIMT programming and GPU memory hierarchy.
• Contributions to open-source projects such as Triton or ROCm are a plus.

Responsibilities

• Design, research, implement, and optimize high-performance Triton/Gluon kernels for large-scale AI models.
• Own and productionize kernels within vLLM and SGL; ensure correctness and throughput.
• Collaborate with compiler engineers to develop the ROCm/LLVM AMDGPU backend across CDNA architectures.
• Drive kernel-level optimizations across memory hierarchy, wavefront execution, and scheduling.
• Profile, microbenchmark, and root-cause bottlenecks in memory bandwidth and latency.
• Debug end-to-end across PyTorch, vLLM/SGL, Triton IR/MLIR, ROCm runtime, and LLVM AMDGPU backend.
• Contribute to open-source Triton, LLVM, and ROCm ecosystems.

Skills

GPU kernel development

Compiler backends

Performance engineering

Triton

SIMT programming

GPU architecture

Education

Bachelor’s or Master’s degree in CS/CE/EE

Tools

Triton

Torch

vLLM

SGLang

IREE

MLIR

LLVM

ROCm

Job description

AMD is seeking a highly skilled software engineer to design, implement, and optimize Triton/Gluon kernels for ML workloads on AMD Instinct accelerators. You will collaborate with research, compiler, and hardware teams to co-design high-performance solutions and push the limits of GPU performance.

You will influence end-to-end performance from kernel coding to runtime integration, contribute to open-source Triton and ROCm ecosystems, and mentor other engineers in SIMT programming and memory