Software Engineer, Compute (GPU)
Fluidstack3 months ago
Base Salary
$208k - $269k/yr
Responsibilities
- Own compute fleet health, metrics pipelines, alerting, and unified GPU health views across Kubernetes and bare metal.
- Automate compute failure detection, triage, parts management, repair, and return-to-service workflows.
- Design and expand GPU qualification platforms covering burn-in, performance baselining, and new product introduction execution.
- Build and maintain Redfish and BMC tooling for firmware-level telemetry, log collection, and low-level fleet access.
- Own the reliability, scalability, operations, and incident response discipline of the production compute fleet.
- Write postmortems and address systemic causes of incidents.
Requirements
- Experience shipping production automation that other teams depend on and comfort working in any programming language with AI coding tools.
- Ability to reason about hardware failure modes at the firmware and silicon level.
- Comfort carrying a pager, running incidents, writing postmortems, and resolving systemic causes.
- Ability to work independently in ambiguous domains and quickly develop competence in unfamiliar areas.
- Fluency with AI tooling, including LLM APIs, MCP servers, agentic frameworks, Claude Code, Cursor, or similar tools.
- Preferred experience with hardware lifecycle management, RMA automation, BMC/Redfish or IPMI tooling, GPU qualification or burn-in frameworks, Temporal or Cadence, Prometheus or Grafana, and Go or Python.
Benefits
- Competitive total compensation package including salary and equity.
- Retirement or pension plan in line with local norms.
- Health, dental, and vision insurance.
- Generous paid time off policy in line with local norms.
Tech Stack
About Fluidstack
Fluidstack provides on-demand GPU cloud infrastructure for AI training and inference, offering dedicated clusters of NVIDIA H200 and GB200 systems. The company sells compute capacity and managed clusters to AI labs, enterprises, and public-sector teams that need high-throughput workloads. Founded in 2017 and headquartered in New York, it is privately held.