Big newbie having questions about GPU computing

Hello everybody!!

I’m new in GPU computing, and i have some questions about compile.

So i downloaded a few days ago all the libraries needed to progamm in CUDA. So i took my time to read the documentation which comes with donwloaded application, but there is only one thing that i did not understand : how do i compile? As i’ve understood, i must use nvcc, but where should i place the .cu files to be compiled, at what command should i type, to obtain the .exe ? Is it “nvcc -cuda” ?

I would also like to now why my board is not detected by CUDA? I run on a FX5200. Could it be because of the drivers?

So i hope you guys will help me because, beacause i’m really blocked.

You need a G80-class GPU to run cuda, an FX5200 is not going to work.

To compile, nvcc will generate an executable a.out.
If you want to generate an executable with a different name “name_exec” ,
nvcc -o name_exec is the right command

Thats really sad. I wanted so to develop a little in CUDA. Will there be any support of the 7-th, 6-th and 5-th series ?

I have another question. Where must i put files that should be compiled? In the same folder than nvcc.exe ?

CUDA requires a very different organization of GPU hardware than was used in the past, so it is almost certainly impossible to support anything before the G80. If you want to try out GPU computing on your existing hardware, I would suggest taking a look at BrookGPU:

BrookGPU runs on older ATI and NVIDIA cards. While not as flexible or as well-documented as CUDA, BrookGPU does let you experiment with data-parallel processing on a graphics card. I used BrookGPU on the GeForce 7 series before getting a 8800 GTX and switching to CUDA.

Your other option is to wait for the next release of the CUDA toolkit and purchase one of the lower end GeForce 8 cards, like the 8500 or the 8600. (Cheapest I’ve found is $100) The current toolkit only works on the GeForce 8800 GTS/GTX, and the Quadro 4600/5600, so don’t run out and buy anything just yet… :)

So as you said i did took i look on it, and really, i prefere CUDA. It seems easier. At least i think so.

I agry. I wanted to buy my self a 8800GTS 640 Mb, but at the same time, why not a 8800GTX. You see, if a change the GPU, i will need also to change the mother board, CPU, memory. To sum up, i will get my self i brand new computer. For the moment i cant’ afford it. So i will wait a little. I will surely try BrookGPU. I don’t have any choices.

Thank you for your help.

[EDIT] :

Can’t i emulate my applications? Even having a GeForce FX 5200? At least this way i will develop in cuda

You can compile CUDA programs in emulation mode, but the emulation is not very realistic. Code that runs correctly in emulation mode may not run correctly on the real card.

It is a reasonable way to get started, but just beware that your code might not work the first time on the real card. :)