I am looking for a way to find the falsification of a matrix so that it can be done on a GPU. SVD / QRs are difficult to parallel and are not supported but MATLAB's GPU, but it seems that LU, though can be run in parallel, is not supported by the GPU of matap. I compared the display and seems to be slower than running on a core CPU.
I'm looking for a pseudo-inverse (or a regular inverse for square matrix) which I can use. According to multibillion, by using mldivide () eliminates the Gauss that applies to GPUs.
I tried to use 'A' but unfortunately it did not run efficiently on the GPU.
Can anyone direct me in a customized code for Elim or Gaussian Elimination?
I have heard about the mega package, but it takes a lot of work to install and compile it and I really need this simple thing.
A C ++ code is also welcome.
All GPUArray in MATLAB R2013a, LU, QR and SVD are supported on GPU. Here is a list of supported functions: linear algebraic functions of GPUArray are implemented using all.
A \ b can run efficiently on GPU, see this example:. As the size of the problem increases, you can expect a decisive powerful GPU (like Tesla) to run faster than the CPU 5 times faster than the CPU.
No comments:
Post a Comment