This project was initiated and led by Dr Vincenzo Brachetta and carried out by PhD researcher Ritesh Moon at the University of Birmingham. Conducted between January and February 2026, it evaluated the performance and practical usability of Julia on BlueBEAR, the University’s institutional high-performance computing (HPC) platform.
BlueBEAR supports multi-core CPU workloads, distributed execution across nodes, and GPU-accelerated computation, with resources managed via the SLURM scheduler. The project developed a set of representative benchmarks to characterise Julia’s performance across these modes and to provide a useful reference for HPC researchers.
Julia is increasingly recognised for combining high-level programming with performance comparable to lower-level languages, reducing the need to switch between languages as computational demands increase. The benchmarks were designed using a progressive approach: single-core execution was used to establish baseline performance and to highlight practical considerations such as just-in-time compilation overhead and the need for warm-up runs. Subsequent stages explored shared-memory multi-threading, distributed multi-process execution, and GPU acceleration using CUDA.jl.
The results demonstrate that Julia integrates well with BlueBEAR and aligns with SLURM-based resource allocation. While scaling performance depends on both hardware and workload characteristics, Julia provides a coherent pathway from prototyping to larger-scale execution within a single language. This project establishes a foundation for broader adoption of Julia on BlueBEAR, with future work focusing on extending benchmarks to domain-specific workloads and developing best-practice guidance for reproducible and scalable HPC workflows.ing best-practice guidance for reproducible and scalable HPC workflows.