HPC/AI Materials Discovery Performance Engineer

CuspAI; Amsterdam; Hybrid; EUR 110,000 - 150,000; Full time

Benefits offered by this job

Equity in CuspAI

28 days holiday (DE, NL, UK)

26 weeks parental leave (primary)

Professional development budget

Impactful work on materials discovery

Job summary

CuspAI is seeking a Performance Engineer to bridge frontier AI models and high‑performance computing infrastructure. You will profile training runs, identify bottlenecks, and help design low‑level optimizations to enable larger, faster models for chemistry and physics.

In this role you will collaborate with AI researchers and the infrastructure team, develop real-time dashboards for MFU, memory bandwidth, and cluster uptime, and drive cost-efficient use of NVIDIA accelerators to advance

Qualifications

• 3+ years in performance engineering or HPC/AI workloads.
• Expert-level GPU knowledge and CUDA kernel optimization experience.
• Deep understanding of distributed training and networking.
• Proficient in profiling tools for performance tracking.
• Ability to work autonomously in a fast-paced startup.

Responsibilities

• Analyze model architectures to map operations onto hardware, focusing on NVIDIA GPU kernels and NCCL.
• Implement efficiency techniques like quantization, pruning, and distillation to reduce latency.
• Streamline data pipelines to keep accelerators data-fed during large training runs.
• Build real-time dashboards and observability for MFU, memory bandwidth, and cluster uptime.

Skills

GPU architecture

CUDA kernel optimization

Triton

Distributed training

Profiling tools

Autonomy

Tools

Nsight Systems

Nsight Compute

Weights & Biases

Job description

CuspAI is seeking a Performance Engineer to bridge frontier AI models and high‑performance computing infrastructure. You will profile training runs, identify bottlenecks, and help design low‑level optimizations to enable larger, faster models for chemistry and physics.

In this role you will collaborate with AI researchers and the infrastructure team, develop real-time dashboards for MFU, memory bandwidth, and cluster uptime, and drive cost-efficient use of NVIDIA accelerators to advance