OpenCL for GPU computing has been flying under the radar ever since CUDA became popular and the myth spread that it would more efficient. With the three major vendors, Nvidia, AMD and Intel, now all providing powerful data-center GPUs in the high-performance computing sector, it becomes more important than ever to be able to write GPU software once and deploy it on all supercomputers without wasting any time on porting.  In this workshop, I will introduce OpenCL and highlight the significant benefits that OpenCL offers over other proprietary languages, in particular it’s excellent compatibility across all GPU hardware since 2009. On a brief case study, I will teach the main techniques of OpenCL performance optimization up to the physical roofline model limit. The workshop will conclude with an open discussion, where participants can seek advice for their own applications.

Target Audience and Prerequisites: Researchers using HPC; basic programming experience in C/C++ would be good.

Instructor: Dr. Moritz Lehmann

Agenda:

1. Introduction to OpenCL
2. GPU performance optimization techniques
3. Multi-GPU parallelization
4. Case-study
5. Open discussion

Registration: To participate in this webinar and receive all the information you need, we encourage you to register using the provided link. As the event will be conducted remotely, registered participants will receive instructions on how to access the webinar online.