This course will give a short introduction to NVidia GPUs and Intel Xeon Phis, which are currently the two most popular accelerator architectures. After a short hardware overview we will present two simple pragma-based programming models for accelerators: offloading for Xeon Phi systems and OpenACC for GPUs. Some basic performance considerations and limitations due to the memory traffic between the host and accelerator will be discussed. Last part of the course will give a short one-day introduction to the lower level approach for GPU programming using CUDA.AgendaTuesday 28th October

09:00 - 09:15 Course introduction
09:15 - 10:15 Introduction to Xeon Phi
10:15 - 10:30 Coffee break
10:30 - 11:30 Offload
11:30 - 12:00 Exercises
12:00 - 12:45 Lunch break
12:45 - 13:45 Data access and performance
13:45 - 14:30 Exercises
14:30 - 14:45 Coffee break
14:45 - 15:45 Exercises
15:45 - 16:15 Advanced topics and wrap-up
Wednesday 29th October

09:00 - 09:30 Introduction to GPUs
09:30 - 10:30 OpenACC basics
10:30 - 10:45 Coffee break
10:45 - 12:00 Exercises
12:00 - 12:45 Lunch break
12:45 - 13:45 Data access and performance
13:45 - 14:30 Exercises
14:30 - 14:45 Coffee break
14:45 - 15:45 Exercises
15:45 - 16:15 Advanced topics and wrap-up
Thursday 30th October

09:00 - 10:15 CUDA programming I
10:15 - 10:30 Coffee break
10:30 - 11:15 Exercises
11:15 - 12:15 CUDA programming II
12:15 - 13:00 Lunch break
13:00 - 14:00 Exercises
14:00 - 14:30 CUDA programming III
14:30 - 14:45 Coffee break
14:45 - 15:45 Exercises
15:45 - 16:15 Course wrap-up
Learning outcome
The students should know the basics of accelerators and utilize them using either offloading or OpenACC.Pre-requisites
Good knowledge on C programming language is needed for the CUDA part. Some basic knowledge about parallel programming is beneficial, but no prior knowledge on GPUs is needed.Price: Free of chargeLecturers: Sami Ilvonen (CSC), Mikko Byckling (CSC)

https://events.prace-ri.eu/event/295/