HPC Network Engineer, InfiniBand and RoCE

You build the fabrics that let GPUs train together across nodes. You design InfiniBand and RoCE networks with rail-optimized topologies, bring them up on site and test them until collective traffic runs at the rate the build calls for.

Team: GPU cluster deployment, Kind: Full-time, Location: On site at client locations in the United States, with travel, Pay: $128,000 to $178,000 a year

Duties

  • Design InfiniBand and RoCE fabrics for GPU clusters, including rail-optimized and fat-tree topologies.
  • Plan switch counts, optics and cable lengths with the cabling technicians before each build.
  • Configure and run subnet managers, partitions and routing on InfiniBand fabrics.
  • Set up RoCE on Ethernet switches with lossless settings, congestion control and quality of service.
  • Bring up each fabric on site, and fix link errors, faulty optics and miscabling.
  • Test each fabric with bandwidth, latency and collective communication tests across all nodes.
  • Watch fabric health and error counters, and replace failing links before they slow training.
  • Document the topology and settings, and hand them to the team that runs the cluster.

What you bring

  • 3+ years building or running high-performance networks for HPC or GPU clusters.
  • Hands-on InfiniBand administration, including subnet managers and fabric diagnostics.
  • RoCE configuration on data center switches, including PFC and ECN.
  • Linux networking and RDMA tools.
  • Scripting in Python or Bash.

Nice to have

  • An NVIDIA networking certification.
  • NVIDIA UFM or another fabric management tool.
  • NCCL tuning for multi-node training.
  • Optics and cables for 400G and 800G links.

Pay and benefits

$128,000 to $178,000 a year

Full-time W-2 employment with health insurance, a retirement plan and paid leave.

No application deadline. We accept applications for this role on an ongoing basis.

Location

On site at client locations in the United States, with travel

Apply

We review applications as each team grows and contact you when there is a match.

Rillor is an equal opportunity employer. We consider qualified applicants without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, age, disability, veteran status or any other status protected by law.

If you need an accommodation to apply, email careers@rillor.com.

Notices

Trademarks. NVIDIA, AMD, Ceph, Kubernetes and related product names are trademarks of their owners and are used only to identify products. No affiliation or endorsement is implied.