This workshop introduces participants to GPU computing techniques in the context of Computational Fluid Dynamics (CFD), using T-Flows, an open-source CFD code available on GitHub. The focus will be on the practical challenges and strategies involved in porting classical CPU-based code to GPU architectures. Participants will gain a conceptual and technical understanding of:

• Why and how to adapt CFD code for GPUs.
• Key architectural differences between CPUs and GPUs.
• Hardware and software requirements for effective GPU computing.

‍

• Target Audience: Researchers, graduate students, and engineers interested in high-performance computing (HPC), CFD, and code development/optimisation for GPU architectures. Prior experience with CFD and programming (C++/Fortran) is recommended but not mandatory.
• Attendees should bring a laptop. Access to GPU-enabled computing infrastructure and all necessary software will be provided during the workshop.

By the end of the workshop, participants will:

• Explore the GPU-accelerated version of T-Flows.
• Learn how core CFD routines were modified for GPU use.
• Run selected cases on provided GPU-enabled systems.

‍

Program

Hour 1: Introduction to GPU Computing for CFD (Theory)

• 15     min — Welcome, objectives, and overview of the workshop
• 20     min — Introduction to GPU architecture vs. CPU:
   
  • Core differences
   
  • Strengths and limitations of GPUs in CFD
 
• 25     min — Why port CFD code to GPUs?
   
  • Performance motivations
   
  • Memory bandwidth, parallelism, compute capability

Hour 2: Adapting CFD Code for GPU (Concepts + T-Flows Background)

• 15     min — Overview of T-Flows architecture (brief intro to solver, structure,     and numerical methods)
• 25     min — GPU porting strategies:
   
  • Identifying performance bottlenecks in CPU code
   
  • Managing memory (global, shared, constant)
   
  • Choosing between CUDA, OpenACC, OpenMP offloading, etc.
 
• 20     min — Specific GPU modifications in T-Flows:
   
  • Which routines were ported
   
  • Data structure and loop parallelization strategies

Hour 3: Hands-On Session – Running T-Flows on GPU

• 10     min — Setup and orientation: access to provided GPU-enabled environment
• 20     min — Compiling and running a basic T-Flows case (GPU vs. CPU)
• 30     min — Guided exploration of GPU kernels:
   
  • Profiling performance
   
  • Observing speedups and resource usage

Hour 4: Advanced Topics + Q&A

• 20     min — Open discussion and Q&A:
   
  • Challenges in porting CFD to GPU
   
  • Extending GPU support in T-Flows

Participant questions