Questions about using CUDALink, Mathematica's method to accelerate programs using NVidia graphics cards.

learn more… | top users | synonyms (1)

7
votes
0answers
643 views

Poor CUDALink performance, when calling a kernel multiple times

When working on a project of mine I noticed that CUDALink shows very poor performance when calling CUDA kernels multiple times in a loop-construct. To demonstrate this I put together this little ...
6
votes
0answers
247 views

Does OpenCL/Cuda memory leak if not Unloaded? Problem with allocating/loading GPU memory in Module

The GPU memory manager documentation is not too clear, does GPU memory leak if not explicitly unloaded or if the symbol that holds the reference to the GPU memory gets collected, the GPU memory gets ...
5
votes
0answers
426 views

How to invert a matrix with 100 trillion elements?

I have a $10^{7}\times 10^{7}$ matrix with the following properties: it is sparse it is gamma-5 it is almost Hermitian. If $M$ is the matrix we are trying to invert, we know $M=A M^{\dagger} A$ ...
4
votes
0answers
101 views

errors in CUDA example

It seems that there are some errors in the documentation page Consider these functions in the documentation CUDA Programming. However, I'm not a expert in CUDA nor C, could you check whether these are ...
4
votes
0answers
105 views

A simple experiment to understand CUDAFunctionLoad

In an attempt to answer my own question (Beginner CUDA: two dimensional blocks and two dimensional threads?). Please excuse the long post, it contains for examples I constructed as an experiment to ...
4
votes
0answers
295 views

Beginner CUDA: two dimensional blocks and two dimensional threads?

This is a follow up question to: CUDA: setting grid dimensions. The purpose of the question is to understand how Mathematica is interfacing with CUDA's architecture. I have a related question/request ...
3
votes
0answers
72 views

Figure out how to copy to constant memory (CUDA)

I found a function. The help gives: CUDALink`Internal`CUDAConstantMemoryCopy[mem1, mem2] copies constant memory to and from host memory. As of now I have not ...
3
votes
0answers
112 views

Getting CUDALink (Toolkit 7.5) to work on Linux, Mathematica 10.0.2, GTX750

I want to make use of some CUDA functionality on my machine. It is running Gentoo Linux (64bit), Mathematica 10.0.2 on an Intel Xeon E5-1620v3 with a Geforce GTX750. So everything on the hardware/OS ...
3
votes
0answers
55 views

Monitor CUDA function

Is there a way to monitor CUDA function without bringing much performance reduction. For example, Map can be monitored by ...
3
votes
0answers
192 views

Calling CUDA cuFFT

Following ybeltukov's excellent answers (see here and here) about how to access the CUDA library functions and interact with Mathematica through CUDAMemory, I tried to compile the code given in ...
2
votes
0answers
80 views

calling CUDA library functions using MathematicaGPULibrary (WGL)

Recently I want implement a sparse linear solver in Mathematica with cuda, and I found the following resource: calling CUDA library functions. As it shows, I tried to use CreateLibrary function to ...
2
votes
0answers
100 views

Looking for a working mathematica CUDA port of NVIDIA's nbody.cu

I am trying to extend an example that is provided as one of NVIDIA's prototypical examples: http://docs.nvidia.com/cuda/cuda-samples/index.html#cuda-n-body-simulation I'm trying, but haven't been ...
2
votes
0answers
942 views

Can Mathematica automatically take advantage of the GPU on the new Mac Pro?

I'm upgrading my Mac to one of the new Mac Pro R2D2 Darth Vader models that have just been released. They come standard with twin AMD FirePro D300 cards, or twin D500 cards. I don't need that much ...
1
vote
0answers
243 views

Can GPU (CUDA) speed up LinearSolve?

Is there a way to use GPU (CUDA) to speed up LinearSolve?
1
vote
0answers
103 views

Why does the iteration not work here in cuda?

I tried to implement Jacobi method for 1D Poisson's equation using CUDALink. The kernel is (I use only 1 block): ...
1
vote
0answers
272 views

CUDA* commands are very slow on OS X and Mac

I have Mathematica 8.0.4.0 installed on my new MacBook Pro running OS X 10.8. I also have the latest CUDA drivers (and even the CUDA SDK) from nVIDIA. I've noticed that commands like CUDAQ[] take a ...
0
votes
0answers
27 views

CUDATranspose bad result

A fairly simple experiment, I am trasponsing an identitiy (3x3) matrix, and I got a weird result. I tried also CUDADot and works as expected. The code I used: ...
0
votes
0answers
79 views

CUDA functions cannot be compiled

i5-3210M, GT 630M; Windows 8.1 x64, Visual Studio 2013; CUDA 6.5 has been installed. One of the clues I found is that GT 630M didn't appear in the list of graphics cards supported by CUDALink which ...