I want to define a very large matrix with about 5000*5000 elements, and calculate it using cuda on gpu. The volumen of an array matrix is limited, so usually I used lapack++ library for c/c++ or cvMatrix in openCV, but there is some problem to use them in cuda(I will write it in another topic). Does someone know, how can I define a large matrix in cuda without any other library? With this quantum is that necessary, the large matrix to partition and directly to define as more small array matrix?
5000x5000 isn’t large by GPU standards - that is only 100Mb in single precision or 200Mb in double precision. Standard GT200b GPUs come with at least 896Mb of memory, and doing linear algebra with that size matrices is easy. The NVIDIA BLAS, CUBLAS, contains functions that can define and manipulate matrices of that size without writing a single line of code for the GPU.
thx for ur always helpful answers!
can u suggest me a standard “cu” lib for image processing?
best regards
Sorry, no. I don’t do image processing with CUDA.