วีดีโอ

Great video thx! but min 5:50 first row miss one 4 and last row should be 0 11. Is it correct ?

Thank you! The comment stated at time 10:42 is probably one of the most important parts! Essentially, one thread (within a given grid) will create/allocate the shared array and all other threads within the same grid will ignore that line of code and assume that it has been handled by the first thread.

Good job :)

excellent explanation, easy to catch up all the concepts and process.. cool

This is an incredibly helpful series. Thank you!

5 years later & this is the best training I've found so far. Thank you Josh!

Smooth one right here

Still great cuda content 5 years later. Thanks for the lectures!

i have cudaitis

Typo at 3:41? The slide says that the device is the device plus the host, that should just be for heterogeneous right?

Extremely helpful! Thank you for the good lecture :)

hey! terrific videos on CUDA... would be great if you could continue uploading follow up videos:)

I am running this kernel to test: __global__ void IncrementCuda(int * d_out, int N) { int i = threadIdx.x + blockIdx.x * blockDim.x; int tmp = -1; if(i < N) { tmp = d_out[(i+1)%N]; //__syncthreads(); d_out[i%N] = tmp; //__syncthreads(); } } Notice I have commented out the __syncthreads(). Calling with N = (up 20000000) , generates perfectly sequenced blocks. I would expect that I would see some 'jags' and gaps in the output i the absence of synchronization. I see none. I have also experimented with calling or not calling cudaDeviceSynchronize(). Again, no change. What is happening? How do I force a race condition in CUDA and observe it's effects?

One of the best conceptual overviews of any ML-related topic, not just CUDA. 4 years later - still worth watching. Props

By far the Best Explanation... Thank you sir

Please teach such things more often and make a playlist on youtube Sir ... things like openmp, embedded C, gnu toolchain, etc. are very rarely taught well in TH-cam.

Great content indeed!!! I'm willing to pay for such content 🔥

You definitely deserve atleast 1M subs and vivers! great stuff, keep doing good work !!! Subscribed!

I got to say, not coming from C) that void** confuses the hell out of me.

very good introduction! Thanks.

implementation of the kernel will be left as an exercise lmao. now out to buy an Nvidia gpu!

please show how to make multiple different custom elements on the same page

this was so helpful, feel like I need to do more than just subscribe 😀

Thank you from the bottom of my heart for those lectures!

CUDA 101

Great series. Thanks. Subscribed. Cheers

@Josh where is the rest!

6:52 // says to cpy Host to device, fcn says to cpy Host to device. btw great playlist!

Very simple and elegant thanks!

extremly good explanation of complex topic in an easy way.. You are awsome

why are you using a double pointer at 6:27? why not just d_c? why &d_c?

th-cam.com/video/4APkMJdiudU/w-d-xo.html Here you mention the threads don't execute at the same time though they perform the same operation. Is that because sometimes there are data dependencies between the threads? You say its because they are working on different data, (the value of data shouldn't matter) right?

I also really enjoyed this series of videos and would be interested in more [advanced ones] on this topic =).

I really like your videos. Are you going to record the following videos on more advanced use of GPU/Cuda in practice? That would be really great. I mean mostly optimization. Best, Michał

you taught me way more than I learned in 2 weeks of GPU programming lectures 😭. THANKS A BUNCH🙏🏽

I enjoyed it.

🙌 "Promo sm"

Thanks! I was looking for a quick intro to custom elements and this was great :D

The best CUDA tutorial !!!

Super valuable series thanks for making it! 4 years too late but still never late to say thank you :)

Great stuff! Everything I needed to form a decent mental model of a GPU

Hello Sir! I have a doubt ... wouldn't the transfer of data from cpu to gpu (global memory) be the bottle-neck in the entire architecture? You teach great! Thank You !!!

yo i cannot begin to describe the gratefullness from this playlist amazing short formed and beneficial

Excellent series. Very clear and accurate lectures. This is the best CUDA lectures I have ever seen. Academia, please keep up the good work. You are well qualified to be educators.

Non sono solito a lasciare commenti, ma in questo caso non posso farne a meno. Ho seguito l'intera serie di video su CUDA e non posso che ringraziarla per la semplicità e la chiarezza con cui spiega i vari argomenti!

cant understand your english ascent..😢

where can I get the slides? Thanks

2:43 what does it mean by’ because it is executed on many data elements, the memory access latency can be hidden with calculations instead of big cache memory.’?

Wonderful tutorial!So,how to optimize cuda program to fully utilize?

Thank you sir for the great lectures, how can we access the rest of the course, is there more than those 6 parts?