Senior AMD GPU Infra Engineer for ROCm AI Stack

Evergrid; New York (NY); Hybrid; USD 180,000 - 220,000; Full time

Benefits offered by this job

Health insurance

Paid time off

Home office stipend

Job summary

Evergrid seeks an experienced HPC GPU operations engineer to own the health, reliability, and performance of AMD GPU clusters. You will lead Linux systems, kernel debugging, and ROCm-based ML stack across production-scale AI workloads.

You will be on-call for outages, optimize GPU topology, and work with data centers and vendors. A strong background in HPC, Python, Bash, and ROCm is required.

Qualifications

• 5+ years in HPC or GPU cluster operations.
• BS/MS in CS/CE/EE or related field.
• Hands-on with AMD MI-series GPUs and driver/kernel debugging.
• Strong Linux internals, kernel modules, and performance tuning.
• Experience securing production infra: VPNs, firewalls, SSH, identity systems.
• Proficient in Bash and Python for automation.
• Familiarity with ROCm stack (PyTorch ROCm, JAX ROCm, RCCL, MIOpen).
• Experience debugging high-performance networking and RDMA.

Responsibilities

• Own system health and on-call response for outages and GPU failures.
• Be the primary point of contact for POC and active customers' GPU clusters.
• Diagnose and resolve issues to minimize downtime and uphold SLA.
• Monitor GPU health, thermals, PCIe topology, memory errors, cluster load.
• Coordinate repairs with data centers and hardware vendors.

Skills

HPC GPU clusters

Linux systems engineering

Python automation

Bash scripting

ROCm / ROCm stack

AMD MI-series GPUs

Kernel debugging

On-call incident response

Education

Bachelor's or Master's in CS/CE/EE

Tools

InfiniBand/RoCE

SSH/VPNs

Job description

Evergrid seeks an experienced HPC GPU operations engineer to own the health, reliability, and performance of AMD GPU clusters. You will lead Linux systems, kernel debugging, and ROCm-based ML stack across production-scale AI workloads.

You will be on-call for outages, optimize GPU topology, and work with data centers and vendors. A strong background in HPC, Python, Bash, and ROCm is required.