About the job
NVIDIA is at the forefront of the generative AI revolution, building the software and systems that power the world’s most advanced large language model workloads. We are looking for a Software Engineer focused on bring-up, triage, benchmarking, analysis, and optimization of distributed training and inference workloads across NVIDIA GPU platforms at the largest scales we run.
Responsibilities
Bring up, validate, and debug large-scale AI clusters, infrastructure, and end-to-end workloads.
Bring up, tune, and benchmark AI pre-training, post-training, and inference workloads using PyTorch, NeMo / Megatron, TensorRT-LLM, and adjacent NVIDIA AI software stacks.
Perform root-cause analysis of failures in large distributed environments
Contribute to the resilience and failure-attribution tooling that detects, triages, and attributes node, fabric, and workload failures across the cluster.
Build and maintain repeatable benchmark suites, automation, acceptance criteria, and qualification workflows on new platforms.
Tune runtime settings, communication parameters, and deployment configurations in close partnership with framework, systems, and platform teams.
Deliver actionable, data-driven recommendations based on profiling, benchmark results, and cluster characterization.
Qualifications
Minimum
Bachelor’s or Master’s in Computer Science or a related technical field (or equivalent experience).
3+ years of experience developing software for AI, HPC, or systems-level applications.
Hands-on experience with multi-GPU or multi-node workloads and CUDA-aware distributed execution.
Background with debugging and scaling distributed systems.
Experience debugging and triaging AI applications across the full stack, from the application level toward the hardware.
Experience operating workloads in scheduled, containerized cluster environments.
Excellent analytical, debugging, and communication skills, and a collaborative approach across teams.
Strong Python and C/C++ programming skills.
Preferred
Hands-on experience with NCCL and CUDA-aware distributed execution.
Deep familiarity with the RDMA software stack (NCCL, IB verbs, UCX, libfabric) and with InfiniBand / RoCE congestion debugging.
Experience building acceptance tests, benchmark harnesses, regression gates, or cluster qualification tooling for AI platforms, including MLPerf.
Experience diagnosing performance jitter.
Experience building resilience, fault-detection, or failure-attribution systems for datacenter-scale infrastructure.