Design and bring up high-speed GPU cluster fabrics: InfiniBand (Quantum) and Ethernet (Spectrum-X/RoCE).
Fabric engineers make sure thousands of GPUs talk to each other at line rate.
GIRIVON (everything.support) supplies AI and high-performance computing infrastructure in India — GPUs, AI servers, workstations, networking, storage, components and accessories — together with the engineering services to design, deploy and run it for enterprises, universities, research labs and data centres.
What you will do
- Design rail-optimised fat-tree topologies, cabling plans and optics bills of materials
- Bring up InfiniBand subnets (UFM/OpenSM) and RoCE fabrics with lossless configuration
- Validate with ib_write_bw, NCCL all-reduce and congestion tests
- Troubleshoot link errors, cabling faults and performance regressions
What we are looking for
- 5+ years in data-centre networking, including InfiniBand or RoCE in production
- Strong understanding of fat-tree topologies, ECN/PFC and adaptive routing
What you get
- Statutory benefits as applicable: Provident Fund, gratuity, paid leave and public holidays
- Work-related travel reimbursed
- Hands-on work with current-generation GPU, server and networking hardware
Skills
- InfiniBand
- RoCE
- Spectrum-X
- UFM
- NCCL
- Optics and cabling
Hiring process
- Apply with your Google Drive résumé link below — you get a reference number immediately.
- The hiring team reviews every application and contacts shortlisted candidates by email or phone.
- Interviews: a screening call, a technical or role interview, and a final conversation. Field and technician roles include a practical assessment.
- Offer with the salary and joining date in writing. GIRIVON never asks candidates for money at any stage.