Lead GPU Architect for Next-Gen AI Accelerators

EngineersOfAI; Milpitas (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Graphcore is seeking a highly accomplished GPU Architect to define the next generation of AI accelerators and multi-GPU cluster architecture. You will lead the technology characterization, reliability, and interconnect performance strategies that ensure our compute fabrics scale flawlessly.

You will collaborate deeply across hardware, firmware, and AI silicon teams to build GPU infrastructure capable of pushing the absolute limits of parallel processing and hardware efficiency.

Qualifications

• 10+ years of GPU, AI accelerator, or highly parallel system experience.
• Strong microarchitecture expertise including SIMD/SIMT and scheduling.
• Manufacturing knowledge for AI compute units and rack-level L11 cooling solutions.
• Experience with RDMA interconnects and hardware clustering.
• Proficiency in C++/Python for performance modeling and workload profiling.

Responsibilities

• Hardware-Software Co-Design to accelerate ML frameworks and localized inference engines.
• Performance Modeling with cycle-accurate simulators and analytical models.
• Influence long-term silicon architecture roadmaps with GPU SoC teams; mentor engineers; drive standards from feasibility to tape-out.
• Reliability analysis including MTBF, FIT rates, IEFR, IFR, and lifecycle curves.
• NPI manufacturing knowledge to detect and correct manufacturing frameworks affecting datacenter products.

Skills

GPU architecture

Microarchitecture

Systems Interconnects

Programming (C++, Python)

Analytical reasoning

Education

BS or MS or equivalent

Tools

RDMA networks

Performance modeling tools

Job description

Graphcore is seeking a highly accomplished GPU Architect to define the next generation of AI accelerators and multi-GPU cluster architecture. You will lead the technology characterization, reliability, and interconnect performance strategies that ensure our compute fabrics scale flawlessly.

You will collaborate deeply across hardware, firmware, and AI silicon teams to build GPU infrastructure capable of pushing the absolute limits of parallel processing and hardware efficiency.