Don't expect to get much out of this course. The material is full of errors, both trivial and technical. It is abundantly clear that quality was not a first order concern when this course was created. You'll get significantly more correct information perusing other CUDA learning resources.

That said, the course has one saving grace and that is that for labs and assignments you get access to Coursera's hosted programming environments which in this case include a GPU instance that you can use to learn on your own. Those hosted environments work really well - kudos to the Rhyme team for that!

Assignments are badly defined, almost everyone doing this course is clueless. The lectures are mediocre. Not sure what I learnt from this course, not much GPU programming.

Not a bad course, but it could be so much better. On the positive side, I learned quite a bit about CUDA programming. Although I have dabbled with it in the past - learning on my own - a structured course introduced me to areas such as using the different types of memory effectively.

However, I found the course very trying at times. First, some of the lessons appear out of order in the overview window on the left. On several occasions, discussion of a programming assignment appeared AFTER the programming assignment. If you take this course, watch out for that. Sometimes, the discussions are useful.

My second major criticism is in the content of the lessons themselves. On very few occasions is there any discussion of the objective of the programming assignment. Not only does one have to figure out how to do the assignment, one has to figure surmise the objective of the assignment, i.e., "what is this program supposed to do?". This is not clear at all in may cases. Obviously, one can figure all this out after some very unnecessary frustration - I did - but it's so unnecessary and should not be a factor in a course offered by a major institution such as Johns Hopkins.

Finally, I saw very little attention from the teacher or an assistant to questions that appeared in the discussion forums. The teacher answered a couple of questions, but most go unanswered. From the sparsity of the discussions, I suspect that very few people take this course.

I'm going to look around for other sources of training before continuing with the other two courses. If I can find it elsewhere, I will certainly go there first.

The worst course I've seen

it is the best course I was ever taken, but the study had some of the most challenging assignments. I am challenged to fix it though, however, it needs to be fixed. most of the people were unenrolled because of this.

I good course that gives you hands on experience with cuda GPU programming. You have a VS code desktop in the browser with a LINUX terminal. LInux machine is connected to a nvidia A10 GPU. very cool.

Require enhancements in terms of video and slides contents

Not clear about the assignments

the assignment is unclear

- The course is a good follow up to the previous course, which introduced the learners to the basics of concurrent programming.

- The quality of content in this course however is not up to the mark. Though the labs do give an opportunity of getting hands on experience to the learners, I often found the lab codes to be incorrect or filled with bugs. The lab of gray-scale image had an incorrect file reading function, both for CPU and the GPU producing incorrect image data as compared to the original image)

- Some slides teach incorrect content, such as the allocation of mapped and pinned memory using cuda functions (cudaHostAlloc is written as cudaHostMalloc in the lab code and slides, and the functions cudaMallocHost and cudaHostAlloc actually refer to the same function in the cuda_runtime.h file, which the lecturer says are used for different purposes. Also the syntax of cudaHostGetDevicePointer is incorrect, both in code and course slides). During this section, I often found myself referring to the documentation, which guided me better.

- The content could be helpful to people, but there is no denying that the quality of the content needs to be improved.

- The lecturer's content delivery during the videos could also use some improvement. Throughout the videos I found it hard to keep paying attention to the videos and used other sources to learn certain topics.

ラボ課題で「間違いが含まれるソースコードをノーヒントで修正する」

という内容が多いが、

・出題者による誘導が致命的に下手で、意図が全く理解できない

・主題のGPUプログラミングと関係ない、ただのケアレスミスの修正

・教材全般にナチュラルにケアレスミスが多いので、課題なのかただのミスなのか分からない

唯一の救いは、ほとんど理解できないままある程度までエラーを潰せば

合格点には達することができることだが、

肝心のGPUプログラミングの勉強にはほとんど役に立たない。

とにかく課題がお粗末すぎる。何も考えられていない。

気持ち的には1つ星だが、スタックはしなかったので2つ星にしておく。

I think it is fantastic, I would like to offer my gratitude to the author of this course even though I haven't yet finished it all.

GOOD

good

pro: knowledge structure and assignment

cons: voice not clear

This course requires more experience than you expected Pros: - You don't need to setup CUDA env by your own. They provides online lab with Nvidia CUDA GPU chip and VSCode - Instruction for each programming assignment is fine. Some duplicated between "Lab" vs. "Programming Assigment" session. I'm confuse why they need two separate session - Easy to run a test and submit the result Cons: - This course is extremely poorly organized. The video lecture just really short (3-4 minutes) with just slide deck, students need read a lot of document to be able to complete programming assignment. - The introduction to shared memory is too short, hard to applied it to the real world

I do think that the instructor need to be more interactive with the listener, especially vocally. Also, explaining the material shall be different to the attached references; hence, It can complete each other not be just a repetition in a presentation. The programming assignment is pretty good, honestly. Thanks for that. I was the most important in the class

Júpiter bugou o código primário e infelizmente não consigo reiniciar ou restaurar para versão inicial, mesmo reinstalando.

2nd course in this specialization has lots of error and ambiguity in assignments.

Reasonably useful material, but many slides had errors because of which I had to frequently look at CUDA docs to get the correct info about things like the actual name of functions mentioned in the slides, the correct order of the arguments, etc. The descriptions of the assignments were also really bad and for every single assignment I had to go to the discussion forum to actually understand what the assignment wanted me to do.