GPU Performance Engineer (AI)

TL;DR

GPU Performance Engineer (AI): Optimizing GPU kernels and distributed training communication for high-performance AI workloads with an accent on throughput, latency, and scaling efficiency. Focus on profiling complex bottlenecks, implementing kernel-level tuning, and ensuring stable performance across multi-node clusters.

Company

hirify.global is a specialized engineering firm focused on optimizing high-performance computing and AI infrastructure.

What you will do

• Profile real-world AI workloads to identify compute, memory, and communication bottlenecks.
• Optimize CUDA kernels and tune NCCL collectives for multi-node scaling.
• Implement and validate performance improvements using controlled experiments and benchmarks.
• Reduce performance variance and tail latency in distributed training environments.
• Build and maintain benchmarking suites and regression guards in CI pipelines.
• Collaborate with ML and systems engineers to provide actionable optimization guidance.

Requirements

• 3–7+ years of experience in GPU/HPC performance engineering or CUDA optimization.
• Strong proficiency in C/C++ and CUDA development.
• Deep expertise with profiling tools like Nsight Systems, Nsight Compute, and CUPTI.
• Solid understanding of GPU architecture, memory hierarchy, and synchronization.
• Experience with multi-GPU communication and NCCL troubleshooting.
• Professional proficiency in English for technical documentation and collaboration.

Nice to have

• Experience with PyTorch, TensorFlow, JAX, or Triton.
• Familiarity with RDMA/Infiniband and topology-aware communication.
• Knowledge of mixed precision training and numerical stability trade-offs.