Description

You will optimize end-to-end machine learning training performance, focusing on deep GPU architecture and low-level system tuning.

This role is on-site.

Responsibilities

• Debug and optimize ML training performance across the full stack.
• Profile and tune GPU kernels using PTX, SASS, warps, and Tensor Cores.
• Implement optimizations for CUDA graph launches, warp-level synchronization, and asynchronous memory operations.
• Tune high-performance networking for GPU clusters using InfiniBand, RoCE, GPUDirect, and NVLink.
• Leverage NCCL and MPI for distributed GPU training and collective communication.

Required Skills

• 5+ years of experience in systems performance engineering or ML infrastructure.
• Deep knowledge of GPU architecture including memory hierarchy and cooperative groups.
• Hands-on experience with CUDA GDB, NVIDIA Nsight Systems, and Nsight Compute.
• Proficiency with GPU libraries: Triton, CUTLASS, CUB, Thrust, cuDNN, and cuBLAS.
• Strong understanding of latency and throughput optimization in GPU environments.
• Experience with distributed training frameworks and high-performance networking technologies.
• Bachelor's degree in Computer Science or related field.

Key Skills

Education

Bachelor's degree