Senior Software Engineer, NCCL and CUDA - CSP Engagements

NVIDIA•Redmond, WA

About The Position

NVIDIA is seeking a Senior Software Engineer, NCCL and CUDA specialization to join our Cloud Service Provider (CSP) Engagements team, focusing on ML software stack functionality and performance for datacenter products such as GB300 and Vera Rubin. This role involves working with the customers to understand the functional and performance issues in the libraries layer for deployment at scale. The role combines deep technical expertise in workloads, NCCL and CUDA libraries, frameworks, and system software interaction to solve the customer issues and bring the next generation of innovation.

Requirements

Experience with parallel programming models and with communication libraries (MPI, NCCL, NVSHMEM) run time.
Experience with performance optimization and profiling tools (e.g., Nsight, nvprof).
Excellent C/C++ programming and debugging skills, with experience in CUDA development.
Good exposure to PCIe and NVLINK.
Deep understanding of operating systems and data-center system architecture.
Knowledge of high-performance networking like InfiniBand, and RoCE.
Proficient understanding of compute, networking and cloud deployment, specifically on bare-metal and VMs.
BS or MS in Computer Engineering, Computer Science, or related field (or equivalent experience).
8+ years of system software validation experience.
Ability to communicate effectively and collaborate with partner and customer teams.

Nice To Haves

Strong software architecture experience.
Experience with deep learning workloads training and inferencing.
Experience conducting performance benchmarking and developing tooling on HPC clusters.
Familiarity with containers, cloud provisioning and scheduling tools such as Docker, Kubernetes, SLURM, and Ansible.

Responsibilities

Engage with our CSPs to root cause functional and performance issues in NCCL and CUDA libraries.
Analyze and improve multi-GPU workloads performance through profiling, benchmarking, and tuning.
Understand and solve NCCL and NVSHMEM data movement issues in multi-node clusters.
Understand and solve CUDA porting issues for customer workloads.
Apply datacenter-specific scheduling and topologies for optimal performance.
Debug and resolve complex issues related to GPU computation, memory, and transports.
Collaborate with customers to understand their workload integration specific challenges to NCCL and CUDIA libraries and suggest tailored solutions aligned with the NVIDIA ecosystem.
Collaborate with AE, FAE, and solution architects to deliver integrated customer solutions and technical documentation.
Collaborate with internal teams to help customers use the latest advancements in CUDA and in NCCL.