Any plans to release a C++, Java, .NET, D, Python/Ruby/LUA wrappers for CUDA in the future? Well, the current C SDK is nice, but to add some OOP designed-wrappers won’t hurt!
I’d be interested too. Does anyone already have working CU code that is called from Java?
Python for the GPU:
[url=“PyGPU - Python for the GPU”]http://www.cs.lth.se/home/Calle_Lejdfors/pygpu/[/url]
It was written for the pre-CUDA interface (Cg) but could be worth contacting the author for updates.
There are people invoking CUDA from Java.
Have you seen JCublas?
If you have used it, what do you think? I am only just beginning to look at CUDA…so don’t have the ability to say whether it is any good or not. It certainly runs…
Well, the timing they give for sgemm seems 100% inaccurate. They are claiming to calculate AB + C where A,B,C are each 1000x1000 matrices. AB involves 1 billion floating point multiplications. This is also ignoring the additions, as well as the scalar calculations. The 1 billion floating point multiplications alone would be 1/500th of a second, as the 8800GT has a theoretical rate of about 500 gigaflops. 1/500th of a second is 2ms. They claim to be getting .087162ms on the calculation. Either they implemented something terribly wrong, or they don’t know how to time their functions. Either way, in my mind they lack credibility if they can’t do something as simple as this. I don’t think I would use it unless I could look at the source code and verify it’s accuracy.