How to test OpenMP capabilities for GPU efficiency?

Question

CUDA

Hello!
OpenMP parallel programming technology version 4.0 and higher declares the "transparent" use of graphics computing boards, i.e. automatic connection of the GPU power, if possible, without introducing additional pragmas into the code. Which, it seems, should relieve the end user (programmer) and eliminate the need to master Cuda or other technologies in cases where fine optimization is not needed.
At the moment, I know that a certain performance gain can be achieved when using the compiler from Intel in conjunction with their own processor and graphics card, but this is expensive.
QUESTION: how to test the capabilities of OpenMP for the efficiency of using the computing resources of GPUs in a more budgetary way?
For example, I want to follow this path:
On a computer with a Core i7 processor and a GeForce GT 740 video card (in the specification www.nvidia.ru/object/geforce-gt-740-ru.html#pdpCon... which indicates the presence of 384 Cuda cores ) install Ubuntu 15.10 (which includes gcc 5.2.1, which according to openmp.org/wp/openmp-compilers supports OpenMP 4.0 )
create an example program that demonstrates the possibilities of using the graph board and compile it with the -fopenmp switch.
Accordingly, two sub-questions:
1) are there enough standard drivers in the system for gcc to use cuda of the graphboard kernel, or is it necessary to "feed" the map to the compiler somehow?
2) what simple program can simply demonstrate the advantage of using a graph board?
3) do you need something else?
Thank you!

Answer 1

The easiest way to tell if a graphics chip is being used is to monitor its temperature. It climbs up - it means it is used.
Usage example:
https://parallel-computing.pro/index.php/9-cuda/43...