xAI was first in the world to build a 100k GPU cluster on an ethernet network and then did it again in 92 days, floors, walls and all. We need an engineer with deep experience in RoCEv2 that can develop at hyper scale while optimizing performance and availability. xAI is building at a furious pace with the latest hardware to help people understand the universe. To make the next significant leap forward, we need to own our own destiny by understanding our current network performance and availability and then optimize it to our training models and how we execute customer inference queries. You will spend most of your days deep inside NCCL, building metric dashboards and tweaking configurations to ensure no performance is left on the table. You will help design the next iteration of our backend and front-end networks that will allow us to seamlessly build-out new GPU infrastructure with little to no engineering assistance.
Requirements
Skills
A minimum of 10 years designing and operating large scale networks with 5 years in the ethernet AI/HPC space.
Deep understanding of congestion control on ethernet with Infiniband an added bonus.
Deep understanding of AI training and inference workloads and how they operate on the network. As part of this you are able to use and debug NCCL and potentially commit to the library.
Expertise in creating a portfolio of metrics for performance and operations to optimize the fleet for training and inference traffic.
Experience with Python to automate away repetitive tasks and facilitate your daily job working with and analyzing large sets of data.
Responsibilities
Develop and optimize performance and availability of RoCEv2-based networks at hyper scale.
Spend most days deep inside NCCL, building metric dashboards and tweaking configurations to maximize performance.
Help design the next iteration of backend and front-end networks to support new GPU infrastructure.
Travel to Memphis for building capacity, participate in team on-call rotation, and assist with scaling and maintenance efforts.
Technologies
RoCEv2NCCLPythonEthernetInfiniband
Descubra se seu currículo está pronto para esta vaga
Veja como nossa IA pode otimizar seu currículo e aumentar suas chances de conseguir esta posição.