[INTERNSHIP] AI Performance Engineering Internship — CPU Benchmarking and Optimization

SiPearl is the European fabless designer of sovereign secure high-performance energy-efficient CPUs for HPC AI and data centres. These CPUs will help address strategic challenges in the fields of security defence medical research energy climate and engineering with a reduced environmental footprint.

In June 25 SiPearl completed the design of the most complex CPU ever designed in Europe Rhea1. Featuring 80 Arm Neoverse V1 cores with 61 billion transistors it is currently in production at TSMC. Sipearl CPUs will equip the two first European exascale supercomputers belonging to EuroHPC JU: Rhea1 will be integrated into the JUPITER machine based in Germany and Rhea2 will be part of Alice Recoque in France.

Incubated within the European Processor Initiative (EPI) consortium and seed-funded by the European Union SiPearl employs almost 200 people in :

• France (Maisons-Laffitte Grenoble Massy Sophia Antipolis)

• Spain (Barcelona)

• and Italy (Bologna)

During this internship you will evaluate selected deep learning workloads on SiPearl processors and reference platforms focusing primarily on CPU inference performance on Arm AArch64 with comparisons against x86 CPUs and other relevant architectures. GPU-based measurements may also provide additional reference points where appropriate. Working closely with the Performance Engineering team you will develop reproducible benchmarks analyse performance bottlenecks and investigate optimisation opportunities. Depending on project progress you may also contribute to computational kernel optimisation performance projections for future SiPearl processors or the evaluation of agentic AI workflows on CPU platforms.

As an intern you will be responsible for the following tasks:

• Select and benchmark representative deep learning models using established benchmark suites such as MLPerf where appropriate.

• Evaluate relevant software components in terms of model support correctness stability and ease of deployment.

• Work with frameworks and inference engines such as PyTorch TensorFlow ONNX Runtime vLLM or SGLang. Automate experiments and collect reproducible performance measurements including throughput latency memory usage and CPU scaling while validating model accuracy.

• Analyse the impact of threading CPU affinity NUMA memory placement numerical precision and parallelisation strategies.

• Compare performance across SiPearl AArch64 processors and reference platforms documenting hardware and software configurations.

• Profile workloads to identify computational memory and software bottlenecks.

• Investigate targeted optimisations in computational kernels libraries or runtime configurations primarily for AArch64.

• Contribute to performance estimates for future SiPearl processors documenting assumptions and limitations.

• Depending on project priorities explore agentic AI workflows and evaluate their performance including task completion end-to-end latency and resource usage.

During this internship you will:

• Develop expertise in experimental methodology reproducible benchmarking and CPU performance analysis.

• Strengthen your Python and C/C programming skills for performance engineering.

• Gain practical experience in parallel execution vectorisation and NUMA-aware performance optimisation.

• Deepen your understanding of deep learning inference frameworks and software ecosystems.

• Improve your technical reporting and communication skills through performance studies and recommendations.

Final-year engineering student or Masters degree student specialising in Computer Science Computer Engineering Electronics or a related field.

Strong interest in computer architecture deep learning and software performance optimisation.

Analytical curious and methodical with strong problem-solving skills.

Comfortable designing experiments investigating technical challenges and presenting findings.

Good written and spoken English with the ability to work collaboratively in an international environment.

Practical programming experience in Python and working knowledge of C or C.

Basic understanding of computer architecture and software performance.

Familiarity with deep learning concepts and at least one framework such as PyTorch or TensorFlow.

Good knowledge of Linux environments.

Location -> Maisons-Laffitte / Barcelona