HPC Engineer

The ideal candidate will be responsible for developing high-quality applications. They will also be responsible for designing and implementing testable and scalable code.

Responsibilities

NVIDIA GPU & Stack

• Manage multi-node GPU clusters on H100/A100/H200 with NVLink, NVSwitch, and full driver/firmware lifecycle (CUDA, cuDNN, NCCL, Fabric Manager)
• Tune EFA/InfiniBand interconnects and NCCL collectives for peak all-reduce throughput
• Profile and optimize workloads using DCGM, Nsight Systems, and Nsight Compute

Distributed ML Training

• Configure multi-node training with PyTorch DDP, FSDP, DeepSpeed ZeRO, and Megatron-LM
• Manage parallel storage (Lustre/GPFS) and benchmark training throughput via NCCL tests

Slurm Administration

• Design partitions, QOS, fairshare, and GPU resource configs for multi-team clusters
• Build job templates, Prolog/Epilog hooks, and Pyxis/Enroot container integrations

Inference & Kubernetes

• Deploy NVIDIA NIM microservices and Triton Inference Server on Kubernetes with GPU device plugins and MIG partitioning
• Apply TensorRT-LLM optimizations (INT8/FP8, continuous batching, tensor parallelism) to meet latency SLAs

Qualifications

• Bachelor's degree or equivalent experience in Computer Science or related field