Software Engineer, ML Performance
Core
Optimizing performance of distributed AI training and inference workloads across the entire stack (model, framework, compiler, CUDA, network) for NVIDIA's GPU solutions.
Role type
Senior IC software engineer (ML performance optimization)
Builds
High-performance GPU training and inference systems
Domain
AI / Accelerated Computing / GPU Systems
Deliverable
production ML models
Required skills
GPU workload optimization, distributed systems analysis, performance profiling, CUDA programming, Python, debugging, statistical analysis of fleet-scale data, framework/runtime/compiler interface knowledge
Preferred skills
PyTorch, JAX, Megatron, NCCL, TensorRT-LLM, vLLM, SGLang, Dynamo, counter-factual analysis, building performance-analysis tools
Technologies
CUDA, NCCL, PyTorch, JAX, Megatron, TensorRT-LLM, vLLM, SGLang, Dynamo
Responsibilities
Analyze performance bottlenecks in distributed training/inference across model, framework, data, compiler, and network layers; Define and track performance metrics across GPU fleets; Drive regression analysis by segmenting workloads and isolating causes; Evaluate automated diagnosis tools for accuracy; Collaborate with build partners to implement performance improvements.
Seniority
Senior, hands-on IC
