newbie questions

Hey all,

I am new to Cuda, but excited to get started…lots of opportunities with programming on the GPU. Just some basic questions for you.

(1) Do you think the following link is a good enough of a resource to really get me going and started?
http://www.nvidia.com/object/cuda_develop.html

(2) This question, although still an easy one, is a bit more specific. I have a simulator that is coded up in java…several thousand lines of code spread over a dozen or so java files. I’d like to run this simulator on the GPU to see how things speed up. Is there any way to make this happen with the simulator written in java? Or do i need to completely rewrite the simulator in C, and in the add in the cuda-specific lines where i send the data to be computed over to the GPU, perform the computation(s), and the retrieve the data. I’m just trying to find a way of not having to rewrite this dang simulator…it’ll be basically like starting from scratch.

Thanks in advance for your help.

-Jonathan

Just wanted to add that this simulator is basically doing a TON of simple computations…but it could potentially be hundreds of thousands, and millions, of these same computations at every tick. Obviously, there will be a computational bottleneck on a CPU…and even on a GPU. But with a GPU, as you all know, I have a lot more growing room before getting to that bottleneck.

So back to my original question, is there a way i can somehow test this and use the GPU while still keeping my simulator in java? (and yes, i’m aware that Cudu is basically a C-like language)

Thanks.

-Jonathan

If you rewrite the code that does the computations, build a dll and wrapper so it can be accessed from java, yes. Otherwise no.

Everything, and much more, you need to know to get started is in the programming guide on the page you posted.

Okay,
I’ve followed the instructions on the Quickstart Guide. Drivers, Toolkit and SDK were installed.

The bandwidthtest worked successfully.

I even watched the Particles example. So things looked up.

Then i kept following the instructions to actually BUILD the bandwidthTest program. I opened up the .cu file within MSVS 2008, and clicked Build.

This was the result:

1>------ Build started: Project: bandwidthTest, Configuration: Debug Win32 ------
1>Performing Custom Build Step
1>nvcc fatal : nvcc cannot find a supported cl version. Only MSVC 7.1 and MSVC 8.0 are supported
1>Linking…
1>LINK : fatal error LNK1181: cannot open input file ‘.\Debug\bandwidthTest.obj’
1>Build log was saved at “file://c:\Program Files\NVIDIA Corporation\NVIDIA CUDA SDK\projects\bandwidthTest\Debug\BuildLog.htm”
1>bandwidthTest - 1 error(s), 0 warning(s)
========== Build: 0 succeeded, 1 failed, 0 up-to-date, 0 skipped ==========

After reading about this online, should i assume that i canNOT use MSVS 2008 (which equates to version 9). Must i go find a previous version somewhere in order to Build this and delelop in CUDA?

Thanks.

Yes, or wait for the CUDA 2.1 release, due soon.

I found this series of articles very useful as a simple intro to programming in CUDA.

Dr. Dobbs CUDA tutorial

And have you looked at the CUDA VS2005 Wizard to make a new cuda project?

CUDA VS2005 Wizard

Does anyone know how soon “really soon” is? Like, are we talking a week or so? Months perhaps.?.

Thanks.

I just built a computer, and installed vista on it (looking for punishment…I know). i have MSVS2008 and can get MSVS2005 if necessary, but would prefer to install and use 2008.

But if i must use MSVS2005, I just want to confirm that this will indeed with with CUDA on vista.

Thanks.

This thread says that CUDA 2.1 will be released in beta version by the end of the year. There seems to be no set release date, so who knows. I suppose there’s always a possibility of it getting pushed back to next year. It’s worth mentioning that tmurray works for NVIDIA, so he knows what he’s talking about. I’d take most of the other things said in that thread with a grain of salt.

It works just fine here. Note that if you are running vista64 and want to build 64-bit cuda apps, you need the full version of VS2005, not express.

In the past, when they have started announcing on the forums that the next version of CUDA is “coming soon” it has been anwhere from a couple weeks to a month before release.

Hi all,

I’m finally coming back to this (CUDA) after having given up when dealing with so many issues i had with Visual Studio 2005. The last post talked about 2.1 working with Visual Studio 2008.

Can someone (or better yet several ppl) confirm they are successfully using VS 2008 to program in CUDA.

If so, is there anything special i need to do other than just install VS2008 and have it work properly. Are there any settings that need to be changed?

Or is it as simple as installing VS2008, opening up one of the sample CUDA programs, building it, and running it?

Thanks in advance for your help.

I normally use Visual Studio 2005, but for CUDA I downloaded Visual C++ 2008 Express Edition and it was very easy to set up. Sample projects compiled and ran right out of the box. I was making my own projects in no time.