Mastering Parallel programming with CUDA platform (Udemy.com)

Great Deep Dive into CUDA with Practical C++ Examples This course offers a solid and thorough introduction to CUDA programming using C++. The instructor explains core concepts like memory hierarchy, kernel launches, and parallel thread execution very clearly. I especially appreciated the hands-on examples and real-world applications — they really helped solidify my understanding. The course structure is well-organized, and the pace is good even for someone new to GPU programming. I also liked that it goes beyond the basics and touches on optimization techniques and profiling tools like Nsight. Overall, it’s a fantastic course if you want to get serious about CUDA development with C++, especially for high-performance computing, computer vision, or deep learning. Overall, it’s a fantastic course if you want to get serious about CUDA development with C++ — especially for high-performance computing, computer vision, or deep learning.

Very good explanation of CUDA and GPU architecture, but some concepts are outdated, such as device synchronization within a kernel. Nvidia has also deprecated the nvprof profiling tool in favor of the NSight applications, which provide full feature parity but in a graphical user interface, which the course does not cover. Overall, this is a great introduction into the world of CUDA programming.

I am about to complete the second section of this course, and over all feel, its a good course, but a bit poorly executed. The accent /voice doesn't bother me, these folks are not native English speaker and we have to live with it. However, the slides are not properly designed, a few minor and obvious errors which could be improved. some irritating mouse/keyboard clicking sound every now or then. But the most IMPORTANT of all, some aspects being taught are not 100% correct. For e.g. in section 2 instructor implements vector reduction in various ways to reduce warp divergence and suggest why one implementation is better than other. (the run time /profiling data backs his claim that some implementations are indeed better than other). But if you think about it, some of those implementations either never had warp divergence, (it only happens when their is branching, not just conditional statement) or those divergences could have very easily be avoided by making trivial changes. I would suggest keep this course as a good "overall" course, think about these concepts etc, but do confirm these concepts with other resources. Don't blindly trust the smaller finer details. Overall, still a decent good course, especially if you are new to CUDA.

Mr. Liyanage has curated an exemplary GPGPU course utilizing Cuda, establishing it as an indispensable resource for comprehending the fundamental aspects of computer programming with Cuda. Employing straightforward algorithms, he adeptly elucidates intricate concepts in GPU programming. It is noteworthy that the course predominantly revolves around Cuda and NVidia architecture. However, the underlying principles concerning arrays of working times, occupancy, and synchronization can seamlessly translate to AMD hardware. Therefore, its applicability extends beyond NVidia, as one can effortlessly substitute "warps" with "wavefronts" to accommodate AMD systems. While the current course provides a solid foundation, I would like to propose the development of a sequel course by Mr. Liyanage. Such a continuation could delve into the practical implementation of algorithms, specifically in areas like image processing, physics simulations, and machine learning. This enhancement would undoubtedly contribute to the course's comprehensiveness and practical utility. I extend my commendations to Mr. Liyanage for his outstanding work and eagerly anticipate the prospect of a sequel.

All of the sections except section 7 are good. For Section 7, I don't understand what is going on or what are the applications for a parallel scan algorithm. Also in Section 7, it seems incomplete or as if parts of the lessons in Section 7 are deleted. For example, "Parallel Compact Algorithm" clip in Section 7, seems to be just thrown in there with no context and I was just left off at how the CSR format works, but I was never shown the implementation of it at all.

This is a decent course, but it has a lot of shortcomings. Sections 2-6 I felt like I learned a lot of interesting things about CUDA, section 1 was okay, but sections 7 and 8 were clearly incomplete and barely comprehensible. There's also a lot of strangeness with the code on screen vs. the code you are meant to follow along with--they don't have a consistent pattern of either reading complete code, or copying along with the teacher. There are a few 'assignments', but as far as I can tell the teacher just rubber stamps all responses without checking if they are actually correct or not. Should you buy this course? Maybe, if you want your hand held a little bit while learning the basics. If you are a go-getter, you can probably get a solid introduction from other places or by reading the cuda toolkit docs / examples. If you do this course, don't worry about the assignments, and skip the last two sections.

The technical part was good. In general great examples and explanations to understand the programming and memory model. I also liked how you approached the hardware explanation. Unfortunately, I have the feeling there is a lack of practical exercises. It would had been really useful to put in practice the theory learned. I also have the feeling that the Bonus section feels kind of incomplete. It is mentioned that there will be lectures for image processing using CUDA but no lecture so far touches that point. Last but not least, I found way to many grammatical mistakes in the slides. It would be nice to correct that.

Extremely useful so far, excellent use of code examples to demonstrate concepts. Focus on different types of performance metrics particularly helpful. Amount of student assessment drops off after Section 1. Make sure to take notes in the lectures, and run and tinker with the code examples to get the most of the course! Good news for Linux users: there is some Windows bias but it's very light. Despite the note in the requisites section VS is not necessary for the course (so far). Much of the crucial concepts taught are demonstrated using nvprof metrics and other options; this means invoking nvprof from the command line with various flags, which is the same in Linux or Windows.

This is a must have course for someone, who wants to program in CUDA. This course will not go all the way to the end, and the final chapter is just a teaser for a hopefully upcoming course, but this is the place to get all the essentials and it does worth to follow through attentively all the lessons, despite the fact, sometimes the pronounciation fails badly, and most of the timeS the captions are just generated nonsense. All together you SHOULD BUY this course, because the instructor's knowledge and intention is a treasure.

While the Image Processing section had interesting information, it really didn't tie in with CUDA at all. I found myself passively watching videos instead of actively solving problems. The lack of self paced hands on problems to solve make this more of a set of reference videos than a real learning class. I spent more time fiddling with my ide to get everything downloaded and running correctly than I did actually using the sample code. I feel I received value, but wish there had been more.