AI Performance Engineering Intern — CPU Inference Benchmarking

SiPearl; Maisons-Laffitte; On-site; EUR 12,000 - 19,000; Full time

Job summary

SiPearl, a European fabless processor designer, seeks an engineering intern to evaluate deep learning workloads on Arm AArch64 CPUs and reference platforms, benchmarking performance and exploring optimisations. You will work with the Performance Engineering team, using PyTorch, TensorFlow, ONNX Runtime, vLLM, and SGLang to automate experiments and collect reproducible metrics.

Develop expertise in experimental methodology, benchmark reproducibility, and CPU performance analysis while

Qualifications

• Enrolment in final year of engineering or Master’s in CS/Engineering.
• Strong interest in computer architecture, DL, and performance optimization.
• Analytical, curious, and methodical with problem-solving skills.

Responsibilities

• Select and benchmark deep learning workloads on SiPearl processors and reference platforms.
• Evaluate software components for model support, correctness, stability, and deployment ease.
• Automate experiments and collect reproducible performance measurements (throughput, latency, memory, CPU scaling).
• Analyse threading, CPU affinity, NUMA placement, precision, and parallelisation strategies.
• Document hardware/software configurations and compare across platforms.

Skills

Python programming

C/C++ programming

Linux knowledge

Performance analysis

Experiment design

Team collaboration

Education

Final-year engineering student

Master’s in CS/Engineering

Tools

PyTorch

TensorFlow

ONNX Runtime

vLLM

SGLang

Job description

SiPearl, a European fabless processor designer, seeks an engineering intern to evaluate deep learning workloads on Arm AArch64 CPUs and reference platforms, benchmarking performance and exploring optimisations. You will work with the Performance Engineering team, using PyTorch, TensorFlow, ONNX Runtime, vLLM, and SGLang to automate experiments and collect reproducible metrics.

Develop expertise in experimental methodology, benchmark reproducibility, and CPU performance analysis while