At Dublin City University (DCU), SCI-SYM (the centre for Scientific Computing and Complex Systems Modelling)

www.ichec.ie/news/1326896908

2 Days GPU Training Course

Requirements
• General programming experience.
• Basic knowledge of C, C++ and Fortran.
• Basic Linux user experience.
Outcome
Course attendees will learn how to perform fast porting of existing sequential applications
onto NVIDIA GPUs using GPU-enabled math libraries, templated and directive-based
programming. These methods are focused on rapid development, rather than maximum
performance, and may better suite for large applications and limited resources.

Course Plan

Day 1
10:00 – 10:30 Registrations and welcoming (coffee break)
10:30 – 12:00 (APC)
1. From GPU to GPGPU
• Performance and parallelism
• GPU evolution
• Parallel systems: multicore and clustering
2. CUDA programming model
• Key principles
• Understanding GPU capabilities
12:00 – 13:00 (APC)
2. CUDA programming model (cont.)
• Introduction to the concept of threads & blocks
• The memory hierarchy
13:00 – 14:00 – Lunch
14:00 – 15:30 (ICHEC)
3. CUDA libraries
• Introduction to ICHEC and GPU system facilities
• CUBLAS & MAGMA + phiGEMM
• CUSPARSE
• CUFFT
• CURAND
15:30 – 15:45 – Coffee break
15:45 – 17:00 (APC)
4. Thrust
• Transforms and functors
• Placeholders and tuples
• Performance considerations
• Thrust and CUDA/C interoperability
17:00 – 17:30 (APC)
5. Rapid CUDA development with directives
• Concept and OpenACC

Day 2
9:30 - 13:00
6. CUDA Guided practical session (APC + ICHEC assistance)
• Simple CUDA programs: code, compile, run, profile, debug
• Thrust: implement data processing transform, measure performance
14:00 – 16:30 (ICHEC)
7. HMPP and hands-on session
10:30 – 10:45 coffee break
13:00 - 14:00 lunch
15:30 – 15:45 coffee break