Your First CUDA C Program

[@piyushkatariya1040]

Very succinct. The only video you need to watch to understand how to parallelize the CPU code to execute on GPU.

[@Techforcurious]

Thank you for explaining parallel processing using a simple example.

[@nathanbrutal4183]

Quite intriguing! Most of this is above my level at this time, but this seems like a nice reference for developers.

[@TheOtherSteel]

I liked your clear explanations and calm pacing, but the text on the computer screen was so tiny I could not read it. (I guessed what was displayed based on your narration, but I would have preferred closer views of the specific sections of code that were discussed.)

[@NVIDIADeveloper]

Thank you for the feedback - will pass this suggestion to the team for future videos.

[@kartikpodugu]

The original code provided when creating a CUDA project in visual studio uses cudaMalloc() and then copies data to GPU. In the example above, cudaMallocManaged() is used. Does that mean cudaMallocManaged() allows us to things using zeroCopy. ?

[@amber9040]

The possibilities are endless!

[@ddd12343]

"Exception thrown at 0x00007FF76D3EEEE4 in matrix_addition.exe: 0xC0000005: Access violation reading location 0xFFFFFFFFFFFFFFFF" It happens in line where we are specifying value of a[i] in for loop: a[i] = i;

[@danielpotter9629]

How come CUDA headers do not need to be incuded? And what happens if the number of threads is less than the array size?

[@AdoobII25]

if you specify 512 threads instead of 1024 then the vector 'c' will hold the summation values up until the 512th index and the other half of 'c' will be zeros. This is because he is using the thread ID of each thread as an index for the vectors a,b and c.

[@panikoti731]

for me, they do: #include "cuda_runtime.h" #include "device_launch_parameters.h"

[@saigirish3275]

I know that GPU's can perform the same operation very quickly. Can you show the time taken for each code to execute and debug and and is there any performance difference in these two codes?

[@JackPunter2012]

The Cuda installer installs profiler. If you open a command window where the code is compiled to and use the command "nvprof [filemame]" it will give you some pretty nice performance metrics. There may be a way to do this from within visual studio but i dont know, im trying to get the damn thing to compile from within Visuall studio cause im on CUDA 9.1 in VS 2017 havent found the solution yet :|

[@adist821]

When i tried debug GPU version i have problem "Exception thrown: write access violation. a was 0x1110112. If there is a handler for this exception, the program may be safely continued." how can I repair this problem? VS 2015 // CUDA 8.0

[@alonzoc537]

where you using the cuda managed allocation function for it?

[@forero910]

@@alonzoc537 yes

[@forero910]

i have VS 2015 with cuda 8.0, and when i tried debug GPU version the same thing happens, do you know how to solve it?

[@hao-wenluo1145]

Dear: When I run the sample code, it shows that "no kernel image is available for execution on the device", "addWithcuda failed!". I have no idea what's going on. Could you give me some opinions? thanks

[@roylevy178]

It runs for me only when SIZE is defined as 1024. Maybe problem with number of thread blocks?

[@xuancao8204]

error : operand types are incompatible ("int" and "int *") //CPU; bunch of error in GPU code. VS 2019. CUDA 10.1

[@kobilica999]

int * == pointer to integer

[@SachaD88]

In function __global__ void VectorAdd pointers can not point to constant (NO const int *a, const int *b) only int *a and int *b, even if data of a and b does not changed as in example given. int parameter can be passed only by value int n. (NO const int& n) Otherwise cudaDeviceSynchronize returns 0x700 error IllegalAddress.

[@sandeepshetty1589]

is there a way to redefine a static array (not heap allocated, so obviously can not use cudaMallocManaged) in to a unified memory space?

[@benjaminf.3760]

Very good! More please :)

[@jvillemare]

Same thing with my Visual Studio: Why does it have red lines under the cudaFree saying the method doesn't exist, but it compiles? This is a real whacky problem.

[@panikoti731]

#include "cuda_runtime.h" #include "device_launch_parameters.h"

[@AngeloHafner]

Muito obrigado por compartilhar este conhecimento...👏👏👏👏👏👏

[@diegolink77]

When I compile the program its ok but when I try tu run it i have a segmentation fault, i tried many solutions to solve it but its the same problem, I find that the reason is for the initial pointers, they are pointing to a null position or some like that, please someone can help me with this. Thanks

[@saikiranreddy6901]

do we require a system with NVIDIA gpu to run this code ?

[@RafaelBadain]

I believe that programs written with CUDA libraries aims at code optimization for GPU processing, so it doesn't make much sense to me run a program optimized for a non-existent GPU in your system.

[@touisbetterthanpi]

Rafael Badain there are non nividia gpus though

[@aytraf]

"OBVIOUSLY", CUDA is a Nvidia tech, wouldn't work with other GPU's. It's platform-specific.

[@Rocket62AL]

Very good, thanks

[@saisrikanthavadhanula]

The code is running fine when I am running with CPU code , but when I add GPU code and debug its prompts that the program_name.exe has stopped working I am using visual studio 2017 and CUDA version 9.0

[@abdjanshvamdjsj]

Hi Srikant, it is a visual studio issue rather than a CUDA issue. Just choose Release and x64, before building the solution.

[@kashifzaheer7804]

hello gentleman why it does not work for visual studio 2017 and cuda version 10.0. any specific reasons

[@viswanathansankar3789]

@@kashifzaheer7804 yes for me also its not working. I also using the same

[@ChristianWidjaya_CW]

@@kashifzaheer7804 use vs 2019 preview with latest 10.1 + nvidia driver

[@SNIPER-my1vo]

maybe im the only one that cant figure this out but new to running code for gpu how the hell do you convert it to cpu code not a very good demonstration on this it seams like he is using hot keys to switch it and I don't know any help please.

[@Jasterthemaster]

yea he CTRL+V pasted code he had in his clipboard you would have to pause the video and write it all out to have the same code as he pasted

[@nathanbrutal4183]

15,000 subscribers even!

[@AramAzhari]

The latest cuda 9.1 doesn't get compiled on latest visual studio 2017 (_MSR_VER = 1912). even if i edit the config header of default cuda runtime project it still doesn't compile.

[@saikiranreddy6901]

Aram Azhari hiii

[@saikiranreddy6901]

Do I need to have a NVIDIA GPU to run this code ?

[@jcw781]

@@saikiranreddy6901 Yes

[@sabrango]

THx!

[@ioannismichalopoulos4173]

.c files compiles and runs properly while .cu file compiles and when it runs, it produces no output (and no errors)!

[@abdjanshvamdjsj]

The program crashes when run on the CUDA version.

[@panikoti731]

#include "cuda_runtime.h" #include "device_launch_parameters.h"

[@ramyasrigorle2609]

I got assert error

[@sastavideoswala]

First program should have been hello world.

[@MaxCoplan]

This man is making this in 2017 while using an IDE looking like it's from 2007

[@ChristopherGray00]

that's because its not dark mode, dark mode visual studio looks better

[@lancelotxavier9084]

Make CUDA open source!

[@404Negative]

does it works on Radeon ??

[@tovarisch3039]

I doubt that very highly

[@stc2828]

Lol you are on the wrong site

[@404Negative]

Hello Woooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooorld

[@SuhelAhmed-hs1jm]

Abc

[@cheonseok84]

thank you for your video. it is very helpful/ here is the man who teach CUDA in korea. he is AI proffessional lecturer teaching Machine learning, CUDA, Aibril, etx website 'blog.naver.com/cheonseok84/220876018626'

[@kasunchamara7497]

Painful Indian English Accent.... Its Always Feel The Difference

[@regielb1]

You call yourself a "cuda developer" - heaven help us all. You don't know how to do a proper presentation. I can't imagine why people like this "example"

[@dstgre]

Are true CUDA developers presentation experts?