Which CUDA do You recommend?

I have Win 7 Ultimate 64 bit and VC++ 2010 (VS 2010 upgrade from VS 2005 Professional),

  1. which CUDA do You recommend?
    2.Is the system sufficient for CUDA 8.0?

The Install Guide shows VS2010 as deprecated but supported, so should work OK.

I have Nvidia Quadro 4000 with Compute Capability 2.0; It will be work from CUDA 8.0?

Yes, it’s the last toolkit to support CC2.0.

When i use build cuda i still have the same problem :

1>------ Build started: Project: cuda1, Configuration: Debug x64 ------

1>G:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\BuildCustomizations\CUDA 8.0.targets(168,9): error MSB4002: There was a failure retrieving the attributes for parameters in the “SanitizePaths” task. Could not load type ‘Microsoft.Build.Framework.IBuildEngine4’ from assembly ‘Microsoft.Build.Framework, Version=4.0.0.0, Culture=neutral, PublicKeyToken=b03f5f7f11d50a3a’.

1>G:\Program Files (x86)\MSBuild\Microsoft.Cpp\v4.0\BuildCustomizations\CUDA 8.0.targets(168,9): error MSB4060: The “SanitizePaths” task has been declared or used incorrectly, or failed during construction. Check the spelling of the task name and the assembly name.

========== Build: 0 succeeded, 1 failed, 0 up-to-date, 0 skipped ==========

Sorry, I can’t help further, as I’ve no Windows Cuda or VS experience.

I don’t have any experience with this specific error. But to me, this reads like everything is not set up correctly or there’s compatibility issues after all.

1.) Did you follow the steps concerning your software setup mentioned in the linked Install Guide?
2.) Is there any special reason for you to use an OS and IDE that both reached their end of life 5 years ago? I imagine this could lead to problems…

does not create a .exe file

I would first try, whether calling nvcc.exe on your code directly from the command line works.

Then you can start to debug running it from Visual Studio.

Besides the installation and setup, the VS problems could also stem from how you created the solution and project.

it is this code. simple example.

include
#include<cuda.h>
#include<math.h>
#include<cuda_runtime.h>

using namespace std;

global void AddIntsCUDA(int *a, int *b)
{
a[0] += b[0];
}
int main()
{
int a=5,b=9;
int *d_a, *d_b;

cudaMalloc(&d_a, sizeof(int));
cudaMalloc(&d_b, sizeof(int));
cudaMemcpy(d_a, &a, sizeof(int),cudaMemcpyHostToDevice);

cudaMemcpy(d_b, &b, sizeof(int),cudaMemcpyHostToDevice);

AddIntsCUDA<<<1,1>>>(d_a, d_b);

cudaMemcpy(&a, d_a, sizeof(int), cudaMemcpyDeviceToHost);

cout<<“The answer is”<<a<<endl;

cudaFree(d_a);
cudaFree(d_b);

return 0;

}

///but this problem is from ever example (sample)
AddIntsCUDA<<<1,1>>>(d_a, d_b);//problem with this example it : third mark "<"is underline in VC++ 2010 edit.

This is a known issue that IntelliSense does not know about Cuda and underlines <<< and some other things. Just ignore the IntelliSense output for CUDA code.

At least in my Windows setup, I can’t call nvcc.exe on my .cu files just like that in order to compile them. If you install the CUDA Toolkit for usage with Visual Studio, you don’t necessarily have the whole build pipeline set up in a way you can also compile from the console. E. g., in my case, nvcc doesn’t find any C++ compiler to use. The command line VS executes when I launch a CUDA project is over 700 characters long.

Have you started the Visual Studio Console (or called something similar, installed entry in the Start Menu of Windows). It is a normal cmd.exe terminal setting the environment variables so that Visual Studio can be called.

Have you tried to copy and execute the VS command line (copy&paste of the 700 lines)?