Senior Software Engineer – AI Middleware
Cornelis Networks, Inc.7 months ago
Remote, United StatesSenior
Responsibilities
- Design and implement performance-critical features for collective communication library enablement on Cornelis Networks’ fabrics.
- Optimize distributed training performance across multi-node, multi-GPU configurations.
- Improve GPU communication paths, including GPU-direct transfers, IPC, and CPU/GPU synchronization.
- Profile distributed AI workloads and identify bottlenecks across the software and hardware stack.
- Tune PyTorch Distributed, TensorFlow/XLA, JAX, DeepSpeed, and Megatron-LM workloads.
- Develop benchmarks and microbenchmarks aligned with real model performance.
- Contribute upstream to AI communication and distributed training projects.
- Participate in design reviews, code reviews, CI, and long-term maintenance.
- Prototype and validate Ultra Ethernet capabilities for AI collective communication.
- Provide deployment input and performance validation, collaborate with kernel/driver, switch, performance, and systems teams, and support advanced escalations through trace analysis and robust fixes.
Requirements
- 8+ years of experience in high-performance systems programming using C/C++ on Linux.
- Strong experience with GPU communication stacks including CUDA/ROCm and NCCL/RCCL.
- Ability to optimize distributed training performance using profiling and tracing.
- Understanding of collective communication concepts and topology awareness.
- Experience delivering production-quality code.
- Open-source contributions in relevant areas.
- Preferred experience with PyTorch Distributed, DeepSpeed, and Megatron-LM.
- Preferred familiarity with libfabric/OFI, UCX, and RDMA concepts.
- Preferred experience with RoCEv2 and Ultra Ethernet.
- Preferred experience building cluster-scale performance test infrastructure.
Benefits
- Remote position for employees residing within the United States.
- Competitive compensation package including equity, cash, and incentives; base salary is determined by skills, qualifications, experience, and location, with possible performance-based incentives including an annual bonus or sales incentives.
- Medical, dental, vision, disability, life, dependent care flexible spending account, accidental injury, and pet insurance.
- Generous paid holidays, 401(k) with company match, and Open Time Off for regular full-time exempt employees.
- Additional sick time, bonding leave, and pregnancy disability leave.
About Cornelis Networks, Inc.
Cornelis Networks builds high-performance, lossless scale-out interconnects for AI and HPC data centers, combining network adapters, switches, and fabric management software. Its products, based on the Omni-Path architecture, aim to accelerate GPU-, CPU-, and accelerator-based clusters for enterprises and research institutions. Founded in 2020 and headquartered in Wayne, PA, the company is privately held and sells hardware, software, and support to operators of large compute infrastructures.