Hall A in Narvik runs 4,096 H200s for teams like Tensorfield and Helix AI, at 99.995% availability over the last twelve months. We want to keep it that way while the hall doubles, and then do it again in Kiruna.
About the role
You will join the six-person reliability team that owns the platform end to end, from the BMC to the scheduler. It is a hands-on role: you write the automation, you lead the incident, and you walk the aisle to see the tray yourself. When N-02 opens in 2027, you help design its runbooks from the first rack.
What you'll do
- Own the health of 1,024-GPU training pods: detection, triage and recovery for GPU, NVLink and InfiniBand faults.
- Build automation that drains, tests and returns nodes to service without a human in the loop.
- Lead incidents with a calm, written process, and write the post-mortems customers read.
- Shape SLOs, alerting and capacity models with Platform and Operations.
- Join a paid on-call rotation, one week in six, with daytime cover from Tromsø.
What you bring
- Five or more years running production Linux infrastructure, at least two in an SRE or similar role.
- Debugging across the stack: kernel, drivers, networking and storage.
- Fluent Python or Go, and comfort with Ansible, Terraform or similar.
- Experience with Prometheus, Grafana or comparable observability tooling.
- Clear written English. Norwegian is welcome, not required.
Nice to have
- GPU or HPC clusters: Slurm, Kubernetes with the GPU operator, NCCL, DCGM.
- InfiniBand NDR fabrics or 400G Ethernet.
- Time in a physical data centre, not only the cloud.
What we offer
A salary of NOK 920k–1.15M, options in Nordlys and six weeks of holiday. We relocate you and your family to Narvik with up to NOK 90k, and give you a hardware budget, ten learning days a year and eight GPUs of your own for experiments. The team works from the hall office four days a week, ten minutes from the fjord.