Crash24h@Tecnico.ULisboa Summer Course

I decided to enrol in this course because it covered CUDA and OpenMP. CUDA became a prominent framework over this past decade enabling the many exciting applications in many topics of CV, but more recently, driving astonishing results in Deep Learning, by empowering popular frameworks like TensorFlow and Caffe. OpenMP became popular for allowing developers to take full advantage of the now standard multicore architectures, with very minimal effort. Expertise on both constitutes a powerful toolset for developing very efficient implementations of the algorithms I'll be working on, over the course of my PhD.

What I found most useful about the course was that it provided a hands-on introduction on three different types of frameworks which cover most needs, in terms of parallelization, a developer might have. CUDA is great when having to perform vectorized operations over high dimensional data, OpenMP allows to easily fully exploit the multicore architectures, and if the problem is too big to be handled by a single machine, that's where MPI comes in and helps approach it from a multi-machine/cluster perspective.

In this day and age, where parallel computational devices are ubiquitous, this course should be considered mandatory for any entry level developer looking to work in highly intensive computing applications. I you have the opportunity to take it, don't miss the chance.

I could start my report with housing. Housing was great. Location is very close to university and to subway station. Thanks to FUSENET. Lisbon probably the best place for any kind of events. In this city everything brings together: Old historical city, modern science, art, ocean, sports.

I have found this course very useful for me. During crush computing course we were studied OpenMP, MPI and CUDA programming. Lectures was understandable. Especially I could light lectures about CUDA. They were very transparent. But strange thing was that on exercises we were using just ready libraries. And all the our codes was sort of “Dear library could you please parallelize my problem?”. Lectures about MPI was little bit hard to follow but exercises covered most of the gaps which appears during the lecture. All the lecturers were very pleasant, and kind. We were asking a lot of questions and no one was stayed without answer. Unfortunately I haven’t participate in last day workshop due to my plane. But the FDTD method with Yee grid and Flip-frog algorithm is parallelize in code that I using and I already know how it could be done.

In the end I want to Thanks all organizators, lecturers and Fusenet for this great possibility to get so much new knowledge which for sure I will use in the future.