A comparison of GPU offloading techniques
217
Building 20.21
NHR@KIT is hosting an introductory topical course for (future) users of the NHR@KIT infrastructure, including the tier 2 HPC cluster HoreKa.
In this talk we give an overview of various GPU offload techniques. We present methods based on compiler pragmas, programming language extensions and specialized libraries. Each of these approaches has its specific advantages and disadvantages. Therefore, we compare the individual solutions in terms of usability and simplicity, achievable performance, supported compilers and supported hardware.
In the interactive spirit of the course, there will be sufficient time for discussion and exchange on this topic.
The course will take place on-site (in-person session).
No course fees apply. The number of participants is restricted.