Скачать с ютуб CUDA Memory Model в хорошем качестве

CUDA Memory Model

Hello CUDA community, We're happy to share our first online meetup! On January 4th we talked about CUDA memory consistency model. Speaker: Georgy Evtushenko Abstract: The main source of non-deterministic concurrency bugs is a misunderstanding of the memory consistency model. If you don't understand the guarantees given by memory model, you are risking to produce a bug, which might not show itself for a thousand executions. Moreover, if you use memory model guarantees correctly you can get significant performance improvements. During this talk, I'll show an example of over 40% runtime reduction achieved by deeper memory model understanding. I've tried to make the talk rather pragmatic, so you can get a common sense of performance benefits along with safety issues. For this talk, you'll need to have basic knowledge of GPGPU programming.