Computational Scientist and responsible for Julia computing at the Swiss National Supercomputing Centre (CSCS), ETH Zurich
In the Fall Semester 2021 at ETH Zurich, we designed and taught a new course: Solving PDEs in parallel on GPUs with Julia. We present technical and teaching experiences we gained: we look at our tech-stack
ImplictGlobalGrid.jl for GPU-computing; and
IJulia.jl/Jupyter for web, slides, and exercises. We look into the crash-course in Julia, teaching software-engineering (git, CI) and project-based student evaluations.
The accelerating outflow of ice in Antarctica or Greenland due to a warming climate or the geodynamic processes shaping the Earth share common computational challenges: extreme-scale high-performance computing (HPC) which requires the next-generation of numerical models, parallel solvers and supercomputers. We here present a fresh approach to modern HPC and share our experience running Julia on thousands of graphical processing units (GPUs).
We present an efficient approach for writing architecture-agnostic parallel high-performance stencil computations in Julia. Powerful metaprogramming, costless abstractions and multiple dispatch enable writing a single code that is usable for both productive prototyping on a single CPU and for production runs on GPU or CPU workstations or supercomputers. Performance similar to CUDA C is achievable, which is typically a large improvement over reachable performance with
CUDA.jl Array programming.
We present a straightforward approach for distributed parallelization of stencil-based Julia applications on a regular staggered grid using GPUs and CPUs. The approach allows to leverage remote direct memory access and was shown to enable close to ideal weak scaling of real-world applications on thousands of GPUs. The communication performed can be easily hidden behind computation.
We present JustSayIt.jl, a software and high-level API for offline, low latency and secure translation of human speech to computer commands or text, leveraging the Vosk Speech Recognition Toolkit. The API includes an unprecedented, highly generic extension to the Julia programming language, which allows to declare arguments in standard function definitions to be obtainable by voice. As a result, it empowers any programmer to quickly write new commands that take arguments from human voice.