A training run is only as fast as its slowest link. Every Dedicated pod in Hall A gets its own InfiniBand fabric with no neighbours on the spine, and we need someone who loves keeping it that way.
About the role
You will join the Hardware team and own the high-speed fabrics in Narvik, from topology design and cabling plans to firmware rollouts and fault isolation. You will also help design the fabric for N-02 in Kiruna from day one.
What you'll do
- Design and deploy NDR InfiniBand fabrics for pods of 512 to 4,096 GPUs.
- Own the subnet managers, firmware and monitoring for every fabric in the hall.
- Find and fix bad links, flapping ports and congestion before customers notice.
- Plan the move to XDR and the fabrics for the first B300 pods.
- Work with Platform on NCCL performance and topology-aware scheduling.
What you'll bring
- Three or more years running InfiniBand or large 400G Ethernet fabrics.
- Solid understanding of fat-tree and rail-optimised topologies.
- Scripting in Python or similar to automate checks and rollouts.
- Time on a data centre floor, comfortable with optics and cable plans.
- Clear written English. Norwegian is welcome, not required.