Just wondering when it comes out. I have some code that can only be run on 4.0.
Hello
The current version of the CUDA toolkit is 4.0RC2 (release candidate 2). I’m currently working on the update, and doing an internal refactoring for several reasons: NVIDIA has changed some of the API, and some adjustments are necessary therefore. Also, JCuda has „grown“ for a while now, I’m also trying to do some cleanups with the major releases (despite the mantra ‚never change a running system‘ - I don’t want to get stuck in a lava flow…). And last but not least, I’m trying to work towards a better support of asynchronous operations, since they become increasingly important. This is not trivial, apart from the pure diligence for updating, for example CUBLAS to CUBLAS2.
(Apart from that, I also intended an implementation of „JNpp“, which also is on its way, and considered some ‚Thrust‘-like library, if this is feasible, but this will still take some time…)
One difficulty is that I hardly have the chance to really test the new functionalities. I’m still working with a single GeForce 8800 here, and only on a 32bit system, so I can only hope that the Unified Memory and „Peer“ functionalities will not be implemented toooo wrong in the first version. In any case, the first version of JCuda 0.4.0 will only be a ‚release candidate‘ as well, and actually only a ‚BETA‘ version, and will require throughout testing - I hope that the refactoring will not break any existing program, but you never know which undocumented features and side-effects clients may already rely on…
May I ask which new features you are going to use? Some support in testing the new features would be welcome … 
bye
It’s finally out
You mean CUDA 4.0 final? I already noticed this here - two days after JCuda 0.4.0 RC2 was uploaded :twisted:
Hopefully I can soon update to 4.0 (final). The update for JCuda should then not take too long (but depends on what they changed between the RC2 and the final version. In any case, I could not yet really test the new features with RC2, since I’m still stuck with my old, single GeForce 8800, but hopefully until the final release there will already be some feedback about the RC2… )
[QUOTE=Marco13]You mean CUDA 4.0 final? I already noticed this here - two days after JCuda 0.4.0 RC2 was uploaded :twisted:
Hopefully I can soon update to 4.0 (final). The update for JCuda should then not take too long (but depends on what they changed between the RC2 and the final version. In any case, I could not yet really test the new features with RC2, since I’m still stuck with my old, single GeForce 8800, but hopefully until the final release there will already be some feedback about the RC2… )[/QUOTE]
Hey,
well I must admit after the hussle with CUDA 3.2, VS2008 I gave up on using jCUDA.
Instead I used jocl and now I’m a happy owner of heterogenous data parallel algorithm that runs on my GT550M and Core i7 i2630QM 
Pitfully GPU is OpenCL 1.0 and CPU already 1.1 Still jocl is a great product
I’ve no trouble with testing your software. Currently I cannot compile the code (same errors with losing precision, missing GL files). I have access to GT200, GT500 and in a few weeks GT400 type cores. All linux. But I really need a runnable version so I can integrate my cuda code with a java app :D.
Hello
The Linux64bit binaries are already available, and it should not be necessary to compile them manually, except, of course, when you already updated to CUDA 4.0 final. The update for 4.0 final is already on the way. I hoped to be able to finish upload it this week, but there are still some other tasks - but it will certainly be available next week, including some fixes for the precision errors during the compilation.
However, there should not be a problem with missing GL header files. Do you still know the error message for that?
bye
Marco
I did update to 4.0 final…
From memory (I’ll try again tomorrow if necessary), while compiling the JCudaDriverJNI directory a header file from ./GL was missing. When I checked the entire directory was not there.
Oh, btw: many many thanks for creating this library! 
Hello
Have you been using the ‘makefile’, or the CMake files? The ‘makefile’ is most likely out-dated, this might cause some problems. But the CMake files should work for Linux as well.
(Concerning the original ‘makefile’: I’ll probably omit these in future releases, since maintaining and updating them is … a hassle, and CMake is working quite well in general)
bye
Marco
In JCudaDriver_common.hpp:
#if defined (APPLE) || defined(MACOSX)
#include <OpenGL/gl.h>
#else
#include <GL/gl.h>
#endif
#include <cudaGL.h>
But looking at it: I think I’m missing the GL library myself. But it could have also been the cudaGL.h… Human memory failure :twisted:
Concerning the GL header, I did a quick websearch, there’s a lot of information (e.g. http://forums.nvidia.com/index.php?showtopic=30464 ), maybe there’s something helpful among that - at least it seems not to be specifically related to JCuda. If the problem persists anyhow, I can try to have a closer look at this - although I rarely have the chance to use a Linux machine at all (and it’s a Linux32 machine, withhout CUDA card, only used for compiling the libraries…).