The aim of this exercise is to evaluate the current (as on Sept. 22, 2014) state of double precision floating point (DPFP) operation support for CUDA on the 15 inch Macbook Pro Retina.
Hardware : Macbook Pro Retina, 2.3 GHz Intel Core i7, 16 GB 1600 MHz DDR3
Relevant onboard graphics card : Nvidia GT 750M
Nvidia CUDA version : 6.5.18
Operating system : Mac OSX Mavericks 10.9.5
DPFP testing software environment : 1. ‘gpubench’ in Parallel computing toolbox in Matlab 2013b: http://www.mathworks.in/matlabcentral/fileexchange/34080-gpubench 2. CUDA-Z version 0.9.220 SVN: http://cuda-z.sourceforge.net/
Test results :
Conclusions (based on test results provided above):
The Nvidia GT 750M card on the 15” Macbook pro Retina running Mac OS X 10.9.5 is not a great CUDA development/proofing platform if the user is interested mainly in double precision, floating point operations. This finding is in direct contrast to claims such as http://blogs.nvidia.com/blog/2012/06/14/new-macbook-pros-make-for-great-cuda-dev-platforms/
Other relevant details:
The Nvidia GT 750M does not have ECC memory which is supported by the CUDA Device Query which returns ‘Disabled’ for ‘Device has ECC support’.