About PolarGrid
PolarGrid is an early-stage Canadian company that runs managed AI inference. Customers bring their own models, and we run them as fast, reliable production endpoints on GPU servers across many sites. We are a small team of about five engineers.
The role
You will own how new GPU servers become ready for customers: from racks at a datacenter to tested, production-ready nodes. You will define when a node is ready, build the automation that gets it there, and keep the fleet healthy once it is running.
This is a hands-on senior role on a small team. You will work with datacenter partners and their on-site staff, and closely with the engineers who own routing and the network.
The work
We bring up large GPU servers at datacenters we don’t own. Each node has to be provisioned, tested hard and proven healthy before customers use it. Between customers it has to be wiped clean. Once in service, failing hardware has to be caught early and taken out of rotation. The hard part is doing this the same way, every time, across many sites, with a small team.
What you bring
You have brought up multi-GPU servers or clusters for production workloads, including acceptance testing and burn-in.
Experience with high-speed GPU networking such as InfiniBand or RoCE.
You have automated bare-metal provisioning and kept firmware and images consistent across a fleet.
You can diagnose GPU and server hardware faults and decide when to repair or replace.
You have run GPUs inside virtual machines or otherwise isolated customers on shared hardware.
You have worked with remote datacenter staff and been on call for a fleet.
You write scripts and tools in a language such as Python, Go or Bash.
You use AI tools well and check their work.
You do not need ML model experience. What matters is that you have put real GPU hardware into service for other people’s workloads and kept it running.
Source: Polargrid job board (Dover).
Report this job
Wrongly listed? Tell us why.