Sr. Staff GPU Network Software Engineer - RCCL, Open Source
Role details
Job location
Tech stack
Job description
Do you want to develop multi node GPU communication libraries to enable high performance computing and machine learning workloads at Exascale? AMD is searching for talented and motivated engineers to develop Communication libraries as part of the AMD Radeon Open Ecosystem (ROCm). THE PERSON: You are accustomed to working in a dynamic, geographically distributed agile team, where partnership and collaboration are paramount. You possess excellent written and verbal communication skills, strong attention to detail, and the ability to express your work in a clear, cohesive fashion. You are results-oriented and accustomed to tight deadlines and changing priorities. Most importantly, you are constantly thinking of ways to improve performance of software and hardware., * Support AMD's RCCL, an open source, GPU-accelerated communication collective middleware and related technologies
- Design, implement, and test networking features for multi-GPU and multi-node communication libraries.
- Benchmark, profile and optimize code to maximize throughput on single-GPU, multi-GPU and clustered systems
- Deliver high-quality code and documentation following best practices for open source software development
- Work with key technical experts across AMD and with our partners and customers to improve ROCm applications, libraries, and tools
- Deploy the libraries on large clusters and debug complex system level issues that could span across different layers of the software stack: gpu kernel drivers, nic driver etc.
Requirements
Do you have experience in Software development?, Do you have a Master's degree?, * Strong background developing applications and libraries in C, C++, and Python
- Experience working with RoCE(RDMA over Converge Ethernet), Libfabric and InfiniBand
- Experience working with Linux Kerner, Device drivers and network drivers.
- Experience designing and building GPU Networks for Large Scale Clusters
- Experience in collective communication libraries: MPI, RCCL, SHMEM and optimization to scale collective communication to scale distributed systems.
- In-depth knowledge of best-practices in software development, including testing, profiling, debugging, documentation, version control, issue tracking, and planning
- Contributions to open-source libraries and applications
NICE TO HAVE:
- GPU software development using HIP, CUDA, or OpenCL
- Understanding of CPU and GPU architectures and low-level optimization techniques including assembly programming and/or vectorization
ACADEMIC CREDENTIALS:
- B.Sc. or B.Eng. degree in Computer Science, Software Engineering, Electrical Engineering, or equivalent
- Advanced degrees, such as M.Sc., M.Eng., Ph.D. are preferred.
#LI-G11 #LI-HYBRID
