I need to decide which open-source integrated GPU to choose i.e. AMD GPU for FFT computation directly on GPU, see here docs.
I cannot demand CUDA support from AMD GPU because its developing tools are just coming 2016 Q1, see press release here and also because Intel Chipsets of Business PCs (Etc Q75, Q85, …) are not supported by PCI-e 16x required by modern GPUs although their personal PC sisters (B75, …) have the support.My current problem is that I am running out of memory (16 GB/32 GB) in the application and I need to balance the load also to the GPU where AMD GPU can help.
I also need well support in user and business PCs.
To get a well supported Matlab code in both ends would be useful in GPU computation: I only need GPU FFT support with GPU arrays in AMD side in for Matlab.
However, I do not understand if it is possible with current libraries here.How to decide AMD GPU for computation with Matlab’s FFT on GPU?
Answer
I do not completely understand your question so forgive me if I do not answer correctly.
Matlab only supports NVidia CUDA enabled devices for GPU accelerated computing. CUDA is an NVidia langauge/toolset requiring an NVidia graphics card. Unless Mathworks (the company that owns Matlab) decides to support the AMD processor you are out of luck. The link you give for the AMD tool that supports cuda is incorrect – it converts cuda code to a format that runs on the AMD GPU – this is not an implementation of CUDA and will NOT work with Matlab.
If you want to use an AMD GPU then you will have to do a lot more work to make it run with matlab. You will need to write a MEX file which interfaces to the AMD FFT libraries (see: http://developer.amd.com/tools-and-sdks/opencl-zone/acl-amd-compute-libraries/).
Good luck!
Attribution
Source : Link , Question Author : Léo Léopold Hertz 준영 , Answer Author : user487432
