Lead HPC Software Optimization Engineer

Responsibilities

• Develop and optimize GPU kernels to accelerate inference and training of large machine learning models.
• Architect and implement strategies for distributed training and inference across multi-GPU and multi-node environments.
• Identify performance bottlenecks using profiling tools and implement optimizations to maximize hardware utilization.
• Design and implement multi-threaded and synchronized compute techniques for modern GPU architectures.
• Build robust benchmarking and validation infrastructure to assess software performance, reliability, and scalability.
• Produce technical documentation and share architectural patterns and code optimization best practices.

Requirements

• Bachelor's or Master's degree in Computer Engineering, Computer Science, or a related technical field.
• Extensive experience in GPU kernel optimization using C++ (17/20).
• Hands-on expertise with CUDA and low-level GPU programming.
• Strong understanding of distributed AI computing, including multi-GPU, NCCL, and MPI.
• Proficiency with performance profiling tools such as Nsight, VTune, or Perf.
• Experience working with AI frameworks like PyTorch, vLLM, Cutlass, or Kokkos.

Preferred Qualifications

• Experience with GPU kernel development in HIP, PTX, or GPU Assembly.
• Deep knowledge of ML/HPC parallel algorithm design (e.g., GEMMs, attention, reductions).
• Familiarity with ROCm/CUDA software stacks, including runtimes and compilers.
• Practical usage of the LLVM compiler flow and tools.
• Advanced knowledge of neural network model data flow and operations.
• Experience with MoE, quantization, and scaling algorithms.

About the Company

AMD is a global leader in high-performance computing, graphics, and visual computing technologies. Our mission is to build products that accelerate next-generation computing experiences—from AI and data centers to gaming and embedded systems. We strive for execution excellence through a culture of innovation, collaboration, and inclusion.