Scale-Up Ethernet Engineer
FuriosaAI · Hsinchu City, Taiwan
Infrastructureonsite
At a glance
Highlights
- Build Ethernet scale-up fabric letting many AI accelerators work as one.
- Bring up a full AI rack as a complete computer with power and management.
- Validate performance and stability using real LLM inference traffic.
Ideal candidate
A breadth-first systems engineer who sees the rack as one computer and thrives chasing performance and stability issues across hardware, firmware, and software under real AI workloads.
Requirements
- Validate scale-up Ethernet endpoint integration with SoC interconnect, DMA, memory map and firmware
- Configure Ethernet endpoints, NICs and switches and understand NOS behavior under congestion
- Isolate performance and error root causes across hardware, firmware, driver, software and fabric
- Evaluate bandwidth and tail latency with collective communication and LLM workloads
- Build rack management with observability, failure detection and test automation
- Experience with OCP rack hardware, power systems, scale-up standards and AI-agent workflows
Lab profile
About FuriosaAI
Builds high-performance, high-efficiency AI compute for the Inference Era.
What they build
- AI inference accelerators
- High-efficiency AI compute hardware
Known for
- Energy-efficient AI chip architecture
- High-performance AI inference
- AI compiler development
Key facts
- Founded in 2017
- Offices in Korea and Silicon Valley
- Compiler-focused R&D lab in Lisbon
First seen · Last seen