I tried using the CUBLAS-XT library routines on Windows 64bit using Visual Studio 2013 and got no compiler errors, but the code doesn’t work. I could not find any example code of using the XT versions of CUBLAS.
Is there any example CUBLAS-XT code for using double precision gemm() routines?
Also I am trying to do gemm() on matrices that are larger than the memory of the card and I am using a Titan Z card which has 2 GPU’s, so double reasons to use the XT version of the library.
I looked through the example programs that ship with CUDA 7.5, then searched the internet for 20 minutes. I am unable to find any complete example program that demonstrates the use of CUBLAS-XT (on a two-GPU system or otherwise). If such an example program doesn’t exist yet, it seems NVIDIA would want to create one going forward, as dual GPU configurations are becoming more common.
Maybe txbob will be able to provide a pointer to CUBLAS-XT sample code when he comes across this thread.