OpenCL. Incorrect global memory size on 64-bit GPU

Accelerated Computing CUDA CUDA Programming and Performance
1

Excuse me in advance if this is a trivial question, but I couldn’t find the answer to it.

I have access to nVidia GTX Titan X GPU with 12GB of memory. clinfo command returns the correct value of CL_GLOBAL_MEM_SIZE:

Address bits: 64
Global memory size: 12884705280
Name: GeForce GTX TITAN X

I have written own code returning this parameter, but it shows a wrong value:

:::::::::::::::::::::::::::::::::::
Platform: NVIDIA CUDA
CL_DEVICE_NAME: GeForce GTX TITAN X
CL_DEVICE_ADDRESS_BITS: 64
CL_DEVICE_GLOBAL_MEM_SIZE: 4294770688
:::::::::::::::::::::::::::::::::::

The content of the corresponding .cpp file is here, [url]http://pastebin.com/YJPs8agw[/url]. I tried to compile it as

g++ -I/usr/local/cuda-7.5/targets/x86_64-linux/include program.cpp -l OpenCL

and

nvcc -I/usr/local/cuda-7.5/targets/x86_64-linux/include program.cpp -l OpenCL

Looks like it is compiled as 32-bit application.
Could anybody tell, how can I obtain all 12GB of global memory in the output of the program?

Thank you in advance,
Natalia

2

Try changing this:

printf("CL_DEVICE_GLOBAL_MEM_SIZE: %u\n", gs);

to this:

printf("CL_DEVICE_GLOBAL_MEM_SIZE: %lu\n", gs);

That was the only change I had to make to run your program correctly.

Also make sure you are using the latest driver for your GPU.

3

Thank you a lot!

It’s really foolishly… :)