GPU Networking Engineer for Distributed Inference & RDMA

The Consensus; New York (NY); On-site; USD 120,000 - 160,000; Full time

Benefits offered by this job

Competitive compensation, including equity

100% medical, dental, and vision insurance

Flexible PTO policy

Paid parental leave

Fertility and family-building stipend

Company-facilitated 401(k)

Job summary

The Consensus is seeking experienced foundational engineers to optimize networking in AI infrastructures. You will play a crucial role in integrating and improving high-performance networking technologies to enhance distributed AI capabilities.

The role demands deep technical expertise in C++ or Python, a grasp on high-performance networking protocols, and the ability to optimize models for unprecedented performance metrics. Join a pioneering team in shaping the future of AI technology.

Qualifications

• Deep experience with high-performance networking protocols (InfiniBand, RoCE v2) required.
• Fluency in C++ or Python with a deep understanding of modern NVIDIA architectures.
• Ability to debug NVLink topology issues and customize solutions.

Responsibilities

• Integrate RDMA/RoCE/InfiniBand capabilities into the inference stack.
• Optimize distributed inference for efficient communication across models.
• Design tools for visualizing packet flow and effective bandwidth.

Skills

High-performance networking protocols

C++ or Python

Memory hierarchy optimization

Custom communication kernels

Tools

NCCL

NVSHMEM

TensorRT-LLM

GPUDirect Storage

Job description

The Consensus is seeking experienced foundational engineers to optimize networking in AI infrastructures. You will play a crucial role in integrating and improving high-performance networking technologies to enhance distributed AI capabilities.

The role demands deep technical expertise in C++ or Python, a grasp on high-performance networking protocols, and the ability to optimize models for unprecedented performance metrics. Join a pioneering team in shaping the future of AI technology.