I just installed the GPU driver and OpenCL SDK on Windows 7-64. Running the test program that queries for presence of the computer’s GeForce 9800 GT succeeds (as well as select others). However, most of the test programs (including Vector Addition) fails and the following Windows 7 popup appears:
“The program can’t start because cudart.dll is missing from your computer. Try reinstalling the program to fix the problem.”
First of all “the program” that should be reinstalled is not identified. After uninstalling the SDK and then re-installing it, the same error occurs. Is the stupid message telling me to go into an uninstall/re-install hell infinite loop? I see lot’s of people on the Web are having this problem. What has Nvidian done to resolve it?
the cudart.dll is in your cuda\bin or cuda\bin64 folder. when you install cuda it creates 2 environment variables
Eg
CUDA_BIN_PATH=D:\GPUComputing\CUDA\bin
CUDA_LIB_PATH=D:\GPUComputing\CUDA\lib
or if you installed 64bit
CUDA_BIN_PATH=D:\GPUComputing\CUDA\bin64
CUDA_LIB_PATH=D:\GPUComputing\CUDA\lib64
So when you are compiling you need to see the target you have in the editor, ie x64 or x86. If it is x86 and you installed cuda64 then it wont find cudart.dll. You will need to compile it as x64.
what i did was i first installed 32bit cuda, copied the bin and lib folder to another directory, unstalled cuda 32, then installed cuda 64 and copied the contents of the bin/lib folder back. So now i can compile both 32/64bit cuda. all i need to do is before opening the editor change the values of CUDA_BIN_PATH and CUDA_LIB_PATH to the target i want to build
At this point, I am not even trying to build anything. I am just trying to run the pre-built test programs that came with the distribution. I searched all directories in the whole system and there is only one copy of cudart.dll and that is in the CyberLink directory structure that I presume was already there before I installed the OpenCL SDK. There is no “\GPUComputing\CUDA” directory structure anywhere in the system.
Perhaps the SDK failed to install correctly because I do not have Microsoft Visual Studio.
paul there is no “GPUComputing” directory, it was my customized folder. the cuda directory is present only if you install cuda toolkiit. and that def has the cudart.dll (cuda-run time dll) if you are trying to run the files from the sdk you donot need visual studio.
To run openCL programs from the NVIDIA GPU Computing SDK Browser, you donot need cudart.dll. When you open the browser you have 3 tabs
C for cuda samples | openCL samples | DirectCompute samples
For openCL you need to access the programs under “openCL samples”. But if you run “C for cuda samples” and did not install cuda then you get that message.
cuda is different and openCL is different. although for nvidia cards openGL beneath is naively implemented under cuda. For openCL you only need to install the openCL driver if your current driver doesn’t work.